Hit chance

On this page

How it works
Implementations
Variants and combinations
History
Related
Games
References
Sources

Hit chance is the probability that an attack connects. The Gameplay Design Patterns wiki calls the pattern Variable Accuracy, “Game mechanics to simulate differences in the difficulty to hit what one aims at”¹. Each game read here compares a random number, or in Fire Emblem the average of two, with a threshold built from the attack’s accuracy and the target’s evasion.

Move accuracyPokémon Red and Blue (1996). An accuracy byte out of 256, so even a 100% move misses 1 time in 256.^2,3,4,5Pokémon Yellow (1998). Red and Blue's accuracy byte, so a 100% move still misses 1 time in 256.⁶Pokémon Gold and Silver (1999). An accuracy byte after stages, less 20 against a BrightPowder, and 255 always hits.⁷Pokémon Crystal (2000). An accuracy byte after stages, less 20 against a BrightPowder, and 255 always hits.⁸Pokémon Ruby and Sapphire (2002). Accuracy in percent times one combined stage ratio, then ability and item modifiers.⁹Pokémon Emerald (2004). Accuracy in percent times one combined stage ratio, checked before the damage.^10,11Pokémon FireRed and LeafGreen (2004). Emerald's percent accuracy, skipped for every move in Pokédude battles and for the player's in the first battle.¹²
Hit rolls less evade rollsFinal Fantasy II (1988). Every dodge the target makes cancels one of the attacker's hits, so an evasive enemy blunts a flurry.^13,14Final Fantasy III (1990). Every dodge the target makes cancels one of the attacker's hits, so an evasive enemy blunts a flurry.¹⁵Final Fantasy IV (1991). Every block the target makes cancels one of the attacker's hits.^16,17
Dodge onlyDragon Warrior (1986). Most monsters dodge one or two of the hero's attacks in 64, a few far more, and the hero never dodges.^18,19,20Dragon Warrior II (1987). Some monsters dodge often, and a hero only rarely.²¹
Hit rate against evadeFinal Fantasy (1987). Most swings land unless the target's evade outweighs the attacker's hit rate.^{22,23,24,25,26}Final Fantasy VII (1997). The attack's hit % and a quarter of dexterity, plus the attacker's Df% less the target's, against 1–100; luck can force a hit, or for a party member a dodge.²⁷
d20 against armor classRogue (1980). A 0–19 roll plus bonuses against 20 less level and armor class: 5% a point.^28,29NetHack (1987). A 1–20 roll below 1 + Luck + level + armor class + bonuses: 5% a point, as in Rogue.^30,31
Two rolls averagedFire Emblem: The Binding Blade (2002). Hit less avoid, against the average of two numbers from 0 to 100, so even a shown 100 can miss.^32,33,34Fire Emblem: The Sacred Stones (2004). Hit less avoid, against the average of two numbers from 0 to 99, so a shown 80 hits 92.2% of the time, computed with the two rolls taken as independent.^35,36,37
Flat chanceColossal Cave Adventure (1976). A dwarf's knife hits 9.5% of the time after the first volley, more once the dwarves are fed.³⁸
Aim scatterDune II (1992). The rocket launcher's and the Deviator's missiles fly to a random point near the target; turrets' missiles don't scatter.³⁹
Dodge, stat against statEarthBound Beginnings (1989). The target dodges from its Fight against the attacker's, 27 in 256 when they match.⁴⁰
Strength picks a result tableZork I (1980). The defender's strength picks a result table, and a 1–9 roll picks miss, stagger, wound, knockout, or kill.^41,42,43
Miss, then a stat dodgeEarthBound (1994). A miss roll in 16 from the weapon, then a dodge of (2 × the target's Speed − the attacker's) in 500.^44,45
Hit, then evadeFinal Fantasy V (1992). An attack lands whole or misses: one roll to hit, then one for the target to dodge.^46,47
Hit times evadeFinal Fantasy VI (1994). Hit rate scaled by the target's evade, and by a bug every attack reads magic evade; a physical attack on a target with Image rolls against the Image instead.⁴⁸

#How it works

threshold = accuracy(attacker, move), adjusted by evasion(target)
if roll(range) passes threshold:
    the attack hits
else:
    it misses; its damage is thrown away or never computed

The games differ in the range of the roll, in how accuracy and evasion combine, and in whether the roll comes before the damage or after it.

	Roll	Threshold	Can a sure hit miss?	Against damage
Dragon Warrior	0–63, against the hero’s attacks only^19,49	the monster dodges below its nibble $d$ ¹⁹	monsters’ attacks never roll⁴⁹	after¹⁹
Final Fantasy	0–200, per swing²³	168 + hit rate − evade²², 1–255	yes: 200 always misses²³	before, per swing²³
Red and Blue	a byte⁴	accuracy byte × two stage ratios³	yes: at most 255 in 256⁴	after⁵⁰
Crystal	a byte⁸	the same, less 20 for a BrightPowder⁸	no: 255 skips the roll⁸	after⁵¹
Emerald	1–100¹⁰	accuracy % × one net stage ratio¹⁰	no, at 100% and even stages¹⁰	before¹¹
The Sacred Stones	the average of two 0–99 numbers³⁵	hit less avoid, within 0–100³⁷	no: 100 beats every average³⁵	before^35,52
Rogue 5.4	0–19, plus bonuses²⁸	at least 20 less level and armor class²⁸	no: at level + armor class + bonus of 20 or more, every roll hits²⁸	before²⁸
NetHack 3.6.7	1–20³⁰	below 1 + Luck + level + armor class + bonuses³⁰	no: a total over 20 always hits³⁰	before³⁰

#Implementations

#Dragon Warrior

Only the hero’s attacks face a dodge roll. Once the damage is set, an awake monster dodges when a roll of 0–63 falls below the low nibble $d$ of its resistance byte, a chance of $d / 64$ ^19,19. Zero damage counts as a failed attack before that roll¹⁸. A sleeping monster never dodges, and spells skip the roll¹⁹.

A monster’s attack has no hit roll; it misses only when it deals 0⁴⁹. The NES manual says Agility lets the hero “more easily avoid the enemy”⁵³. In the code, Agility raises the hero’s defense power and enters the first-turn and run tests, and no roll lets the hero dodge^54,55,56,49.

In the North American Dragon Warrior II, a monster that isn’t asleep still dodges a hero’s hit on a roll of 0–63, against a nibble the disassembly calls its “evade chance / 64”^21,19. Heroes now dodge a monster’s attack 1 time in 64, and the Clothes Hiding adds a second chance of 1 in 8²¹.

#Final Fantasy

A physical attack’s hit chance is 168 + hit rate − evade²² against a defender that can move, held within 1–255²². Dark on the attacker costs 40, and Dark on the defender gives 40²². Each swing rolls 0–200 with RandAX; a roll of 200 always misses, and any other roll at most the hit chance hits²³. Hit rate grows with level and with the weapon’s hit bonus^57,58. Evade rises by 1 whenever a level-up raises Agility²⁵.

Against a sleeping or stunned defender, the attacker’s hit rate and the defender’s evade drop out, leaving the chance at 168, moved only by Dark²⁴. An evasive target is therefore easier to hit asleep, and an accurate attacker harder; Disch flags the branch as possibly a bug²⁴. Run in the ROM for all 256 generator states, one swing at hit chance 172 gives 221 hits and 4 criticals in 256²⁶; with the attacker in Dark it hits 170 times²⁶.

A physical attack would gain 40 hit chance against a matching monster category or element, but no attack can trigger the match⁵⁹.

In the North American release of Final Fantasy VII, a physical attack hits on At% + Dex/4 + own Df% − target Df%, against 1–100²⁷. Against a sleeping target its chance becomes 255²⁷. Final Fantasy instead drops the attacker’s hit rate and the defender’s evade and keeps the chance at 168²⁴.

#Final Fantasy II to VI

Final Fantasy II splits Final Fantasy’s single roll into two sets of rolls¹⁴. The attacker rolls 0–100 once per point of its hit multiplier against its hit %, the defender rolls once per point of its evade multiplier against its evade %, and each evade cancels a hit¹³. Final Fantasy III rolls the same way on 0–99, and halves the hit % for Blind and, except with a harp or bow and arrow, for a back-row attacker or target^15,13. Final Fantasy IV keeps Final Fantasy III’s rolls^17,15.

Final Fantasy V replaces that loop with one hit roll and one evade roll, so an attack either lands or misses^47,46. Final Fantasy VI scales the hit rate by the target’s evade and rolls 0–99 once⁴⁸. Because of a carry-flag bug, every physical attack tests the target’s magic evade (MBlock) instead of Evade; the disassembly’s BUGFIX_EVADE option, credited to Terii Senshi, restores Evade⁴⁸.

#Pokémon Red and Blue

Red and Blue store accuracy as a byte, $⌊ 255 p / 100 ⌋$ for $p$ percent, so 100% is 255². CalcHitChance scales it by the attacker’s accuracy stage and then by the defender’s evasion stage read in reverse, using the same ratios as the other stats, from ×0.25 to ×4³. It truncates after each step, keeps at least 1, and caps the result at 255³.

MoveHitTest then hits when a random byte is below the result, so even a 100% move lands at most 255 in 256⁴. The test runs after the damage is computed⁵⁰. Swift returns before any check and never misses, even against a Pokémon in the air from Fly or underground from Dig; X Accuracy skips only the roll⁵.

Pokémon Yellow keeps MoveHitTest and CalcHitChance unchanged^6,4.

#Pokémon Crystal

Crystal keeps the byte and the two stage steps, with its own accuracy ratios from ×0.33 to ×3⁸. A BrightPowder on the target subtracts 20 from the threshold⁸. An accuracy of 255 now skips the roll, so a 100% move at even stages always hits⁸. A plain damaging move still checks accuracy last⁵¹.

Pokémon Gold and Silver already have the BrightPowder and the sure hit at 255, and Crystal keeps their BattleCommand_CheckHit unchanged^7,8.

#Pokémon Emerald

Emerald works in percent. It nets the attacker’s accuracy stage against the target’s evasion stage, multiplies the move’s accuracy by that one ratio, and misses when a roll from 1 to 100 exceeds the result¹⁰. Compound Eyes, Sand Veil in a sandstorm, Hustle on physical moves, and an evasion item scale the threshold first¹⁰. A plain damaging move now checks accuracy before it computes any damage¹¹.

Pokémon Ruby and Sapphire already have this check, and Emerald keeps their atk01_accuracycheck with renamed constants^9,10. Pokémon FireRed and LeafGreen add one exception to Emerald’s check: every move skips the roll in Pokédude battles, and the player’s moves skip it in the first battle until tutorial flags are set¹².

#EarthBound Beginnings and EarthBound

In the English build of EarthBound Beginnings, the attacker has no accuracy stat⁴⁰. The target dodges a plain attack with a chance of (256 + ΔFight) ÷ 2 − 102, in 256⁴⁰, where ΔFight is its Fight less the attacker’s; that is 27 in 256 at equal Fight⁴⁰. In the US release of EarthBound, a bash first misses at the attacker’s miss rate in 16, and the target then dodges by Speed instead of Fight^44,45.

#Fire Emblem: The Sacred Stones

The Sacred Stones shows one hit percentage in its battle forecast. It is twice the attacker’s skill, plus the weapon’s hit, half the luck, and the weapon triangle’s bonus, less the defender’s avoid of twice attack speed plus terrain and luck, held within 0–100³⁷. The game then averages two random numbers from 0 to 99, rounding down, and the attack hits when the shown hit is greater³⁵. The average bunches near the middle, so a high chance hits more often than shown and a low one less often. Computed from the code, taking the two rolls as independent, a shown 80 hits 92.2% of the time, 50 hits 50.5%, and 20 hits 8.2%³⁶. The crit roll that follows a hit uses one number⁶⁰.

The Binding Blade, read in its Japanese build, already averages two numbers³². Its numbers divide by 655 rather than scale by 100/65,536, so each can reach 100, and a displayed 100 misses 1,296 times in 2³², computed the same way^33,35.

#Variants and combinations

Critical hit. Final Fantasy reads the critical from the hit roll itself, so a critical is always a hit⁶¹.
Multi-hit attack. Final Fantasy rolls every swing; Pokémon rolls once for the whole move^23,62,63,64.

#History

The games read here run from Dragon Quest (Famicom, 1986) to Emerald (2004)^65,66.

Games

Game	Year	How	Note
Pokémon Red and Blue	1996	Move accuracy	an accuracy byte out of 256, so even a 100% move misses 1 time in 256
Pokémon Yellow	1998	Move accuracy	Red and Blue's accuracy byte, so a 100% move still misses 1 time in 256
Pokémon Gold and Silver	1999	Move accuracy	an accuracy byte after stages, less 20 against a BrightPowder, and 255 always hits
Pokémon Crystal	2000	Move accuracy	an accuracy byte after stages, less 20 against a BrightPowder, and 255 always hits
Pokémon Ruby and Sapphire	2002	Move accuracy	accuracy in percent times one combined stage ratio, then ability and item modifiers
Pokémon Emerald	2004	Move accuracy	accuracy in percent times one combined stage ratio, checked before the damage
Pokémon FireRed and LeafGreen	2004	Move accuracy	Emerald's percent accuracy, skipped for every move in Pokédude battles and for the player's in the first battle
Final Fantasy II	1988	Hit rolls less evade rolls	every dodge the target makes cancels one of the attacker's hits, so an evasive enemy blunts a flurry
Final Fantasy III	1990	Hit rolls less evade rolls	every dodge the target makes cancels one of the attacker's hits, so an evasive enemy blunts a flurry
Final Fantasy IV	1991	Hit rolls less evade rolls	every block the target makes cancels one of the attacker's hits
Dragon Warrior	1986	Dodge only	most monsters dodge one or two of the hero's attacks in 64, a few far more, and the hero never dodges
Dragon Warrior II	1987	Dodge only	some monsters dodge often, and a hero only rarely
Final Fantasy	1987	Hit rate against evade	most swings land unless the target's evade outweighs the attacker's hit rate
Final Fantasy VII	1997	Hit rate against evade	the attack's hit % and a quarter of dexterity, plus the attacker's Df% less the target's, against 1–100; luck can force a hit, or for a party member a dodge
Rogue	1980	d20 against armor class	a 0–19 roll plus bonuses against 20 less level and armor class: 5% a point
NetHack	1987	d20 against armor class	a 1–20 roll below 1 + Luck + level + armor class + bonuses: 5% a point, as in Rogue
Fire Emblem: The Binding Blade	2002	Two rolls averaged	hit less avoid, against the average of two numbers from 0 to 100, so even a shown 100 can miss
Fire Emblem: The Sacred Stones	2004	Two rolls averaged	hit less avoid, against the average of two numbers from 0 to 99, so a shown 80 hits 92.2% of the time, computed with the two rolls taken as independent
Colossal Cave Adventure	1976	Flat chance	a dwarf's knife hits 9.5% of the time after the first volley, more once the dwarves are fed
Dune II	1992	Aim scatter	the rocket launcher's and the Deviator's missiles fly to a random point near the target; turrets' missiles don't scatter
EarthBound Beginnings	1989	Dodge, stat against stat	the target dodges from its Fight against the attacker's, 27 in 256 when they match
Zork I	1980	Strength picks a result table	the defender's strength picks a result table, and a 1–9 roll picks miss, stagger, wound, knockout, or kill
EarthBound	1994	Miss, then a stat dodge	a miss roll in 16 from the weapon, then a dodge of (2 × the target's Speed − the attacker's) in 500
Final Fantasy V	1992	Hit, then evade	an attack lands whole or misses: one roll to hit, then one for the target to dodge
Final Fantasy VI	1994	Hit times evade	hit rate scaled by the target's evade, and by a bug every attack reads magic evade; a physical attack on a target with Image rolls against the Image instead

Every claim about this mechanic, by game (77)

Colossal Cave Adventure 1976	In Open Adventure (Adventure 2.5), each dwarf that was already in the hero's location before the dwarves moved throws a knife, which hits if a roll of 0–999 is below 95 × (dflag − 2). The first volley always misses, at dflag 2; it raises dflag to 3, so each later knife hits 9.5% of the time (computed). Feeding a dwarf while food is here adds 2 to dflag, so 28.5% after one feeding (computed). Any hit kills the hero.	Code
Rogue 1980	In Rogue 5.4 an attack hits when `rnd(20)`, a roll of 0 to 19, plus a to-hit bonus is at least 20 less the attacker's level less the defender's armor class. The bonus is the weapon's plus (with a fired missile's launcher plus, and the hero's dexterity rings when the weapon used is the hero's wielded one), the attacker's strength adjustment, and +4 when the target isn't running (asleep or held); a hero defender's armor class is the worn armor's, when the hero wears any, less any rings of protection. Computed, the chance is (level + armor class + bonus) / 20, held within 0 to 100%. Damage is rolled only after a hit.	Code
Rogue 1980	Computed from `swing`, a level-1 Rogue 5.4 hero at strength 16 with a +1 weapon hits a monster of armor class 7 45% of the time while it is running, and 65% while it is asleep or held.	Code
Zork I 1980	Zork I's ZIL source sets the player's fighting strength at 2 + score ÷ 70, rounded down (computed from 2 + score ÷ (350 ÷ (7 − 2))), from 2 at no points to 7 at 350, plus a wound modifier that is 0 when unhurt. A villain's strength is its STRENGTH: 2 for the troll and 5 for the thief. Attacked with its bane, the sword for the troll and the nasty knife for the thief, a villain counts 1 less, but no less than 1; a thief engrossed in a treasure counts at most 2.	Code
	In Zork I's ZIL source each blow, by the player or a villain, reads 9 entries of a result table picked by the defender's strength, and a roll of 1–9 picks one. A defender of strength 1 uses DEF1, starting 0, 1, or 2 entries in for an attacker of strength 1, 2, or 3 and up. A defender of 2 uses DEF2A for an attacker of 1, or DEF2B starting 0, 1, or 2 entries in for 2, 3, or 4 and up. A defender of 3 or more uses the attacker's strength less the defender's, clamped to −2..2: DEF3A starting 0 or 1 entries in for −2 or −1, DEF3B likewise for 0 or 1, and DEF3C for 2. The results are miss, stagger, light wound, serious wound, knockout, and kill; the DEF3 tables hold no knockout or kill, and an unarmed or unconscious villain dies to the player's blow outright.	Code
	Computed from Zork I's ZIL source: at 0 points the player's strength is 2. Swinging the sword at the unhurt troll (2, less 1 for the sword) reads DEF1 from its second entry, so a blow misses 3 times in 9, staggers 2, knocks out 2, and kills 2. Swinging the nasty knife at the thief when he isn't engrossed (5, less 1) reads DEF3A from its start: 5 misses, 2 staggers, and 2 light wounds, and no kill.	Code
Dragon Warrior 1986	Zero damage from the hero ends the turn as a failed attack, before any dodge check.	Code
	After the damage is set, an awake monster dodges when a roll of 0–63 falls below $d$ , the low nibble of its resistance byte, a chance of $d / 64$ ; a sleeping monster never dodges, and spells skip the check.	Code
	A monster's physical attack has no critical branch and no dodge roll for the hero; `EnPhysAttack` goes straight to `EnCalcHitDmg`, and only 0 damage misses.	Code
	The dodge nibble $d$ of the 40 monsters (each dodges $d / 64$ of the hero's attacks, dodge) is 1 or 2 for 27 of them; the Demon Knight's is 15, the Werewolf's 7, the Poltergeist's and Drakeema's 6, six monsters' 4, and the Golem's and both Dragonlord forms' 0.	Code
	The high nibble $h$ of the same byte blocks Hurt and Hurtmore $h / 16$ of the time (hero-spells); it is 0 for 26 of the 40 monsters and 15 for seven, the Metal Slime, Golem, Demon Knight, Wizard, Red Dragon, and both Dragonlord forms.	Code
	Each fight picks one of its zone's five monster slots with equal chance; zone 0, around Tantegel Castle, holds three Slime slots and two Red Slime slots, and zone 1 adds a Drakee, all with dodge 1.	Code
	The Demon Knight, the hardest monster to hit at 15/64, fills one of the five slots in overworld zones 10, 11, and 12; the Werewolf, at 7/64, fills one in zone 12 and in zone 13, which Hauksness also uses, and one on the Dragonlord's castle ground floor.	Code
	The Dragonlord's castle's six sublevels draw from three rows, Werewolf, Green Dragon, Starwyvern, Wizard, and Axe Knight on the first; Wizard, Axe Knight, two Blue Dragons, and Stoneman on the next three; and Wizard, Stoneman, two Armored Knights, and Red Dragon on the last two and the bottom floor. Apart from the Werewolf's 7, their dodge nibbles are 1 or 2.	Code
Dragon Warrior II 1987	In Dragon Warrior II (NA NES), after a hero's normal hit is rolled, a monster that isn't asleep dodges when a roll of 0–63 falls below the high nibble of its second stat byte, which the disassembly calls its "evade chance / 64"; heroes dodge a monster's attack 1 time in 64, and the Clothes Hiding adds a second chance of 1 in 8.	Code
Final Fantasy 1987	A physical attack's hit chance starts at 168; Dark on the attacker costs 40, and Dark on the defender gives 40. Unless the defender is asleep or stunned, it then adds the attacker's hit rate and subtracts the defender's evade. The result is held within 1–255.	Code
	Against a sleeping or stunned defender, base damage rises by a quarter, and the attacker's hit rate and the defender's evade drop out, so the hit chance stays at 168 apart from Dark's 40 either way.	Code
	Each swing rolls 0–200 with `RandAX`; a roll of 200 always misses, and any other roll hits when it is at most the hit chance.	Code
	A physical attack gains 40 hit chance and 4 base damage when its category or element matches the defender's, but neither side of a battle can trigger it: a character's attack reads the character's own category and weakness bytes instead of the weapon's, and a monster's attack is matched against a character's, and all four bytes are always 0.	Code
	Measured in the ROM over all 256 battle-RNG states, one swing with base damage 20, hit rate 10, and crit chance 1 against evade 6 and absorb 4 hits 221 times and is critical 4 times; with the attacker in Dark it hits 170 times, and against a sleeping defender 216 times.	Verified
	A level-up that raises a character's Agility also raises their evade by 1, up to 200.	Code
	A new character's hit rate is 10 as a Fighter, 7 as a Red Mage, and 5 in the other four classes; evade is 53 for the Fighter, Black Belt, and White Mage and 58 for the Thief, Red Mage, and Black Mage.	Code
	Each level-up adds 3 to the hit rate of a Fighter, Black Belt, Knight, or Master, 2 for a Thief, Red Mage, Ninja, or Red Wizard, and 1 for a White or Black Mage or Wizard, up to 200.	Code
	A new game zero-fills every character's stat page, weapon and armor slots included, so the party starts unarmed, with 400 gold.	Code
	The 40 weapons add from 0 to 50 to hit rate: 0 for both nunchucks, the Iron Hammer, and four of the six staves; 10 for early blades such as the Small Knife and Short Sword; 35 for the Defense, CatClaw, Katana, and Xcalber; and 50 for the Masmune.	Code
	Shop 1, the first weapon shop in `lut_ShopData`, sells the Wooden Staff, Small Knife, Wooden Nunchuck, Rapier, and Iron Hammer, with hit bonuses of 0, 10, 0, 5, and 0 and damage of 6, 5, 12, 9, and 9: the most accurate of them, the Small Knife, deals the least damage.	Code
	Shop 1 is the weapon shop in Coneria, the town beside the starting point; Nintendo Power's guide lists Coneria's weapons as shop 1's five.	Magazine
	Across the 128 enemy records, evade runs from 0 (Scum) to 144 (the Air elemental), with a median of 48; 31 records sit below 25, and six reach 100 or more: Air, Nitemare (132), Frost D and Sphinx (120), Wraith (108), and Chaos (100).	Code
	A new game puts the party at overworld tile (153, 165), in land domain $2C, whose eight formation slots all hold 3–5 Imps (evade 6). The domain to its north, $24, gives 3–5 Imps in 48 of 64 battles; its other slots hold 1–3 GrImps (evade 9), 1–2 Wolves (36), a lone Madpony (22), and, 1 time in 64, 3–6 Imps with up to 4 GrImps.	Code
	The Air elemental (evade 144) is in the random battles of five standard maps, $2F, $30, $32, $33, and $3A, as 2–4 or 3–6 of them in one 12-in-64 slot of the last three, to which $32 and $33 add a second, smaller slot of 1–3 (15 and 18 of 64 in all); Wraith (108) fills 42 of 64 chances on map $25, Frost D (120) 30 of 64 on map $34, and Sphinx (120) 24 of 64 on map $0E and 12 of 64 in four overworld domains.	Code
	Nintendo Power's 1990 strategy guide paces the party by level: about 3 for Garland, "Wait until Level 8 or 9" for the Marsh Cave, and "Level 27 or above to have a good chance against Chaos."	Magazine
	Nintendo Power's guide places Wraiths in the Ice Cave and Air elementals in the Sky Castle.	Magazine
	FF1Randomizer names the standard maps by index: $0F, $25, and $26 are the Ice Cave's B1–B3, $2F–$33 the Sky Palace's five floors, $34 the revisited Temple of Fiends' first floor, and $3A and $3B its Air and Chaos floors.	Web
NetHack 1987	In NetHack 3.6.7 the hero's melee attack hits when a roll of 1 to 20 is below a total of 1 + Luck + the strength and dexterity bonus + the target's armor class + the hero's level (a polymorphed hero's monster level) + the hero's to-hit increment from rings of increase accuracy. The total gains 2 against a stunned, a fleeing, or a sleeping target (which the attack wakes), and 4 against one that can't move; it loses 3 while the hero is trapped and more for a heavier load; it adds the wielded weapon's to-hit and the hero's skill bonus, and adjusts for a monk and for an elf against an orc. An engulfed hero always hits. Computed, the chance is (total − 1) / 20, held within 0 to 100%. Damage is computed only after a hit.	Code
NetHack 1987	Computed from both games' code, NetHack 3.6.7's melee roll keeps the rate of Rogue 5.4's `swing`: each point of the attacker's level or the target's armor class adds 5% to the chance to hit, which is (level + armor class + bonuses) / 20 in Rogue and (level + armor class + Luck + bonuses) / 20 in NetHack.	Code
Final Fantasy II 1988	A Fight action rolls 0–100 once per point of the attacker's hit multiplier (adjusted by haste or slow, at least 1) and counts the rolls under its hit %, then rolls once per point of the target's evade multiplier and counts the rolls under its evade %. The evaded rolls are subtracted from the hits, at least 0, and the rest are the hits that deal damage. Blind halves the attacker's hit % and the target's evade %. The check returns no successes when the attacker is a toad or mini, and the disassembly marks as a bug that the evade check reads the attacker's status too.	Code
Final Fantasy II 1988	Final Fantasy rolls once per swing against a single hit chance that already has the defender's evade subtracted; Final Fantasy II splits that into two sets of rolls, the attacker's against its hit % and the defender's against its evade %, and subtracts the defender's successes from the attacker's.	Code
EarthBound Beginnings 1989	In EarthBound Beginnings (English build), the target of a plain attack dodges when a random byte is at most max(0, ⌊(256 + target's Fight − attacker's Fight) ÷ 2⌋ − 102), 27 in 256 at equal Fight; a paralyzed, sleeping, or petrified target can't dodge. The attacker has no accuracy stat of its own.	Code
Final Fantasy III 1990	A Fight action rolls 0–99 once per point of the attacker's attack multiplier and counts the rolls under its hit %; Blind halves the hit %, and when a back-row penalty applies (not for a harp or bow and arrow) an attacker in the back row halves it, and a target in the back row halves it again and adds one evaded hit. The target then rolls once per point of its evade multiplier against its evade %, with no evade rolls when it is a toad, mini, or jumping, and its successes are subtracted from the hits, at least 0. The net hits multiply one damage roll, and none is a miss. Each hand rolls separately.	Code
Final Fantasy IV 1991	A Fight command makes one percentage roll per point of the attacker's attack multiplier and counts the rolls under its hit rate; the target rolls once per point of its defense multiplier and blocks one hit for each roll under its defense rate. The hits left after blocking multiply one damage roll, and none left is a miss.	Code
Final Fantasy IV 1991	Final Fantasy IV keeps Final Fantasy III's physical attack: percentage rolls per point of an attack multiplier, the target's successful evade rolls subtracted, and one damage roll of attack plus 0 to half of it, with the crit bonus added before defense, multiplied by the net hits and capped at 9,999.	Code
Dune II 1992	The rocket launcher's and the Deviator's missiles aim at a random point near the target. 15 times in 16 the miss distance is a random byte halved until it is at most 8 + the distance to the target in whole tiles, and 1 time in 16 at most 8 + a second random byte, in sixteenths of a tile, in a random direction. The rocket turret's missiles carry no such flag and fly to the target point.	Code
Final Fantasy V 1992	A physical attack first rolls 0–99 against its hit % (100 where the attack type calls `SetHit100andTargetEvade`, as fists and katanas do) and misses on a roll at or above it; a target that can dodge then rolls 0–99 against its evade %, and a roll under it is a miss, or a shield block when the target carries a shield.	Code
Final Fantasy V 1992	Final Fantasy IV counts a weapon attack's hits from a loop of percentage rolls, one per point of the attack multiplier, less the target's evade rolls; Final Fantasy V's physical check makes one hit roll and one evade roll, and the attack either lands or misses.	Code
EarthBound 1994	In EarthBound (US release), a bash first misses at the attacker's miss rate in 16: for a party member its weapon's, or 1 with no weapon, plus 8 while crying or nauseous; for an enemy its own. Then the target dodges with a chance of about (2 × target's Speed − attacker's Speed) in 500, never when that is 0 or less, nor while the target is paralyzed, asleep, immobilized, or solidified.	Code
EarthBound 1994	EarthBound (US release) keeps EarthBound Beginnings' order of a plain attack, with the "SMAAAASH!!" rolled before the dodge so it can't be dodged, but changes the numbers: the chance comes from the attacker's Guts alone (about Guts in 500) instead of Fight against Fight; the SMAAAASH deals 4 × Offense − Defense instead of the full Offense; a normal hit deals 2 × Offense − Defense instead of Offense − Defense ÷ 2; and the dodge comes from Speed instead of Fight.	Code
Final Fantasy VI 1994	Unless the target has Image, which a physical attack instead tests with a random byte under $40 (a miss that may remove the Image), an attack that can miss hits when a roll of 0–99 is under its hit rate times the target's stored evade, over 256, where the stored value is 256 − 2 × the evade % (255 for 0). The branch meant to pick Evade for a physical attack tests a carry flag that the check before it has cleared, so every such attack is checked against the target's magic evade (MBlock) and skips the Blind and status adjustments; the disassembly carries a `BUGFIX_EVADE` option, credited to Terii Senshi, that restores Evade.	Code
Pokémon Red and Blue 1996	Red and Blue store a move's accuracy as a byte, $⌊ 255 p / 100 ⌋$ for $p$ percent, so 100% is 255 and 85% is 216.	Code
	`MoveHitTest` lands a move when a random byte is below its accuracy byte, so even a 100% move misses 1 time in 256.	Code
	`CalcHitChance` multiplies the accuracy byte by the attacker's accuracy-stage ratio and then by the defender's evasion-stage ratio, read in reverse; it truncates after each step, keeps at least 1, and caps the result at 255.	Code
	Swift returns from `MoveHitTest` before any check, so it never misses, even against a Pokémon in the air from Fly or underground from Dig; other moves always miss such a target, and X Accuracy skips only the roll.	Code
	A multi-hit move runs the critical test, the damage formula, and the accuracy roll once; every later hit repeats the first hit's damage, and a first hit that misses ends the move.	Code
	Of the 165 moves in `Moves`, Struggle included, 104 are listed at 100% accuracy and 61 below it; the lowest are the three one-hit KO moves at 30%.	Code
	Bulbasaur and Squirtle start with Tackle, 35 power at 95% accuracy, and Charmander with Scratch, 40 power at 100%.	Code
	The strongest special moves of each type trade accuracy for power: Thunder (120, 70%) against Thunderbolt (95, 100%), Blizzard (120, 90%) against Ice Beam (95, 100%), Fire Blast (120, 85%) against Flamethrower (95, 100%), and Hydro Pump (120, 80%) against Surf (95, 100%); the sleep moves run from Sing's 55% to Spore's 100%, and the one-hit KO moves sit at 30%.	Code
	Sand-Attack and Smokescreen (100%), Kinesis (80%), and Flash (70%) lower the target's accuracy by one stage; Double Team and Minimize (100%) raise the user's evasion by one stage. No move changes either stat by two.	Code
	Outside link battles, an enemy's stat-lowering effect first fails when a random byte is below 64, 1 time in 4; the player's moves skip that roll. A move whose main effect lowers a stat, such as Sand-Attack or Smokescreen, then faces the normal accuracy test, while a damaging move's side effect instead takes hold only on a byte below 85, 85 times in 256.	Code
	Pidgey learns Sand-Attack at level 5, and a wild or trainer Pokémon knows the moves its learnset gives up to its level; so Route 1's level-5 Pidgey (slot 9, 3 in 256 encounters), Route 2's (slot 4, 25 in 256), and the rival's level-9 Pidgey on Route 22 can use it.	Code
Final Fantasy VII 1997	In the North American release, `BattleRollPhysicalHit` hits when a roll of 1–100 (`BattleGetRnd164`) is under the hit chance: the attack's hit % (`targetAttackPercent`) plus a quarter of the attacker's dexterity, plus the attacker's Df%, less the target's Df%, cut by 30% under Fury, and raised to at least 1, which never hits. A party member's Df% is a quarter of its dexterity plus its physical evade and an enemy's is its physical evade alone, so a party attacker's dexterity counts twice. The chance is instead 255 against a target under Death, Sleep, Confusion, Stop, Petrify, Manipulate, or Paralysis, for the Death, Nullify, and Absorb affinities and one unnamed affinity flag, and for two flags the code leaves unnamed (`unk234` bit 0, `unk218` bit 5). Before the roll, a 0–99 roll under the attacker's luck/4 makes a sure hit; failing that, when an enemy attacks a party member and the same roll is under the target's luck/4, the attack misses, even at 255. `BattleGetRnd164` scales a 16-bit random number by 99/65,535 and adds 1, so it gives 100 only for $FFFF, and a chance h of 2–100 hits about (h − 1) times in 99 (computed for evenly spread 16-bit values). Final Fantasy VI instead scales the hit rate by the target's stored evade (magic evade, through a bug) and rolls 0–99 against it.	Code
Pokémon Yellow 1998	Yellow keeps Red and Blue's `MoveHitTest` and `CalcHitChance` unchanged, so a move still hits when a random byte is below its scaled accuracy byte, and even a 100% move misses 1 time in 256.	Code
Pokémon Gold and Silver 1999	Gold and Silver's `BattleCommand_CheckHit`, which Crystal keeps unchanged, subtracts a target's BrightPowder value (20) from the stage-scaled accuracy byte, lets a byte of 255 always hit, and otherwise hits on a random byte below it.	Code
Pokémon Crystal 2000	`BattleCommand_CheckHit` scales the accuracy byte by the accuracy and evasion stages, subtracts 20 if the target holds a BrightPowder, and then hits on a random byte below it; an accuracy of 255 always hits.	Code
Pokémon Crystal 2000	`BattleCommand_EndLoop` gives 2–5 hits with Red and Blue's odds and Triple Kick 1–3 hits, 1/3 each, and jumps back to the `critical` command, so every hit rolls its own critical and damage; `checkhit` sits before the loop and runs once.	Code
Fire Emblem: The Binding Blade 2002	The Binding Blade hits when the displayed hit is greater than the average of two random numbers, rounded down; each number divides a 16-bit `NextRN()` by 655, so it runs from 0 to 100, and 100 comes from only 36 of the 65,536 values.	Code
	Computed from the code, taking the two numbers as independent, a displayed 80 in The Binding Blade hits 92.16% of the time, 50 hits 50.44%, and 20 hits 8.19%; a displayed 100 misses 1,296 times in 4,294,967,296, when both numbers come out 100.	Code
	The Binding Blade computes hit as twice the skill, plus the weapon's hit, half the luck, and the weapon triangle's bonus, and avoid as twice the attack speed plus the terrain's avoid and the luck; the hit it rolls against is hit less avoid, held within 0–100.	Code
Pokémon Ruby and Sapphire 2002	Ruby and Sapphire's `atk01_accuracycheck`, which Emerald keeps with renamed constants, scales a move's percent accuracy by one combined stage ratio, applies Compound Eyes (×1.3), Sand Veil in a sandstorm and Hustle on physical moves (×0.8 each), and a target's evasion item, and misses when a roll from 1 to 100 exceeds the result.	Code
Pokémon Ruby and Sapphire 2002	Ruby and Sapphire's `atk8D_setmultihitcounter` gives 2 or 3 hits with probability 3/8 each and 4 or 5 with 1/8 each, the same code as Emerald's; the multi-hit battle script runs `accuracycheck` once and then loops `critcalc`, `damagecalc`, and `adjustnormaldamage` for every hit.	Code
Fire Emblem: The Sacred Stones 2004	In the North American build, an attack hits when the displayed hit is greater than the average of two random numbers from 0 to 99, rounded down; each number scales a 16-bit `NextRN()` by 100/65,536.	Code
	Computed from the code, taking the two numbers as independent and evenly spread, a displayed hit $h$ connects $h (2 h + 1)$ times in 10,000 up to 50 and $10,000 - (100 - h) (199 - 2 h)$ times above it, so a displayed 80 hits 92.2% of the time, 20 hits 8.2%, and 50 hits 50.5%.	Code
	In the North American build, a unit's hit is twice its skill, plus its weapon's hit, half its luck, and the weapon triangle's bonus; its avoid is twice its attack speed plus its terrain's avoid and its luck; the hit shown and rolled is the attacker's hit less the defender's avoid, held within 0–100.	Code
Pokémon Emerald 2004	`Cmd_accuracycheck` combines the attacker's accuracy stage and the target's evasion stage into one stage, multiplies the move's accuracy in percent by that stage's ratio, and misses when a roll from 1 to 100 exceeds the result, so a 100% move at even stages never misses.	Code
	In Emerald a plain damaging move checks accuracy before it computes damage, the reverse of Red, Blue, and Crystal.	Code
	`Cmd_setmultihitcounter` gives 2 or 3 hits with probability 3/8 each and 4 or 5 with 1/8 each; the loop reruns `critcalc` and `damagecalc` for every hit, after one `accuracycheck`.	Code
Pokémon FireRed and LeafGreen 2004	FireRed and LeafGreen's `Cmd_accuracycheck` is Emerald's with one addition: in Pokédude battles every move, on either side, skips the accuracy roll, and in the first battle the player's damaging moves skip it until one tutorial flag is set and the player's status moves until a second.	Code
Pokémon FireRed and LeafGreen 2004	FireRed and LeafGreen's `Cmd_setmultihitcounter` is Emerald's, line for line: 2 or 3 hits at 3/8 each and 4 or 5 at 1/8. The battle script runs `accuracycheck` once, then loops `critcalc`, `damagecalc`, and `adjustnormaldamage` for every hit.	Code
general	The Gameplay Design Patterns wiki defines Variable Accuracy as "Game mechanics to simulate differences in the difficulty to hit what one aims at."	Wiki

References

^The Gameplay Design Patterns wiki defines Variable Accuracy as "Game mechanics to simulate differences in the difficulty to hit what one aims at."WikiGameplay Design Patterns wiki, “Variable Accuracy”
How we know
Game mechanics to simulate differences in the difficulty to hit what one aims at.
lead · archived copy ↗
The quote matches our archived copy.
In the claim ledger
^ a bRed and Blue store a move's accuracy as a byte, $⌊ 255 p / 100 ⌋$ for $p$ percent, so 100% is 255 and 85% is 216.Codepret/pokered
How we know
```
6	db \5 percent ; accuracy
```
data/moves/moves.asm lines 1–8 at af51989 · upstream ↗
```
3DEF percent EQUS "* $ff / 100"
```
macros/data.asm line 3 at af51989 · upstream ↗
Checked against the evidence by its author, an AI model, 2026-10-20.
Checked by a second AI model, 2026-10-21.
In the claim ledger
^ a b c dCalcHitChance multiplies the accuracy byte by the attacker's accuracy-stage ratio and then by the defender's evasion-stage ratio, read in reverse; it truncates after each step, keeps at least 1, and caps the result at 255.Codepret/pokered
How we know
```
5355	ld c, a
⋮
5362	ld c, a
⋮
5366	ld c, a ; c = 14 - EVASIONMOD (this "reflects" the value over 7, so that an increase in the target's evasion
⋮
5400; make sure the result is always at least one
⋮
5413	ld a, $ff ; set the hit chance to 0xFF
```
engine/battle/core.asm lines 5348–5418 at af51989 · upstream ↗
```
1StatModifierRatios:
2; first byte is numerator, second byte is denominator
3	db 25, 100  ; 0.25
⋮
9	db  1,   1  ; 1.00
⋮
15	db  4,   1  ; 4.00
```
data/battle/stat_modifiers.asm lines 1–15 at af51989 · upstream ↗
Checked against the evidence by its author, an AI model, 2026-10-20.
Checked by a second AI model, 2026-10-21.
In the claim ledger
^ a b c d eMoveHitTest lands a move when a random byte is below its accuracy byte, so even a 100% move misses 1 time in 256.Codepret/pokered
How we know
```
5322; if the random number generated is greater than or equal to the scaled accuracy, the move misses
5323; note that this means that even the highest accuracy is still just a 255/256 chance, not 100%
5324	call BattleRandom
5325	cp b
5326	jr nc, .moveMissed
```
engine/battle/core.asm lines 5321–5327 at af51989 · upstream ↗
Checked against the evidence by its author, an AI model, 2026-10-20.
Checked by a second AI model, 2026-10-21.
In the claim ledger
^ a b cSwift returns from MoveHitTest before any check, so it never misses, even against a Pokémon in the air from Fly or underground from Dig; other moves always miss such a target, and X Accuracy skips only the roll.Codepret/pokered
How we know
```
5249	cp SWIFT_EFFECT
5250	ret z ; Swift never misses (this was fixed from the Japanese versions)
⋮
5260	bit INVULNERABLE, [hl]
5261	jp nz, .moveMissed
```
engine/battle/core.asm lines 5247–5261 at af51989 · upstream ↗
```
5289	bit USING_X_ACCURACY, a ; is the player using X Accuracy?
5290	ret nz ; if so, always hit regardless of accuracy/evasion
```
engine/battle/core.asm lines 5287–5290 at af51989 · upstream ↗
Checked against the evidence by its author, an AI model, 2026-10-20.
Checked by a second AI model, 2026-10-21.
In the claim ledger
^ a bYellow keeps Red and Blue's MoveHitTest and CalcHitChance unchanged, so a move still hits when a random byte is below its scaled accuracy byte, and even a 100% move misses 1 time in 256.Codepret/pokeyellow · pret/pokered
How we know
```
5506	call BattleRandom
5507	cp b
5508	jr nc, .moveMissed
```
engine/battle/core.asm lines 5506–5508 at b6696f2 · upstream ↗
engine/battle/core.asm#L5228-L5419 · upstream ↗
Checked against the evidence by its author, an AI model, 2026-10-21.
Checked by a second AI model, 2026-10-21.
In the claim ledger
^ a bGold and Silver's BattleCommand_CheckHit, which Crystal keeps unchanged, subtracts a target's BrightPowder value (20) from the stage-scaled accuracy byte, lets a byte of 255 always hit, and otherwise hits on a random byte below it.Codepret/pokegold · pret/pokecrystal
How we know
```
1595	cp HELD_BRIGHTPOWDER
⋮
1602	sub c
⋮
1609	cp -1
1610	jr z, .Hit
⋮
1612	call BattleRandom
1613	cp b
1614	jr nc, .Miss
```
engine/battle/effect_commands.asm lines 1595–1614 at fc702b6 · upstream ↗
```
15; BRIGHTPOWDER
16	item_attribute 10, HELD_BRIGHTPOWDER, 20, CANT_SELECT, ITEM, ITEMMENU_NOUSE, ITEMMENU_NOUSE
```
data/items/attributes.asm lines 15–16 at fc702b6 · upstream ↗
engine/battle/effect_commands.asm#L1546-L1850 · upstream ↗
Checked against the evidence by its author, an AI model, 2026-10-21.
Checked by a second AI model, 2026-10-21.
In the claim ledger
^ a b c d e f g hBattleCommand_CheckHit scales the accuracy byte by the accuracy and evasion stages, subtracts 20 if the target holds a BrightPowder, and then hits on a random byte below it; an accuracy of 255 always hits.Codepret/pokecrystal
How we know
```
1584.BrightPowder:
⋮
1587	ld a, b
1588	cp HELD_BRIGHTPOWDER
⋮
1594	ld a, b
1595	sub c
⋮
1600.skip_brightpowder
1601	ld a, b
1602	cp -1
1603	jr z, .Hit
⋮
1605	call BattleRandom
1606	cp b
1607	jr nc, .Miss
```
engine/battle/effect_commands.asm lines 1584–1608 at 97187cd · upstream ↗
```
4AccuracyLevelMultipliers:
5	db  33, 100 ; -6 =  33%
⋮
11	db   1,   1 ;  0 = 100%
⋮
17	db   3,   1 ; +6 = 300%
```
data/battle/accuracy_multipliers.asm lines 1–17 at 97187cd · upstream ↗
```
16	item_attribute 10, HELD_BRIGHTPOWDER, 20, CANT_SELECT, ITEM, ITEMMENU_NOUSE, ITEMMENU_NOUSE
```
data/items/attributes.asm lines 15–16 at 97187cd · upstream ↗
Checked against the evidence by its author, an AI model, 2026-10-20.
Checked by a second AI model, 2026-10-21.
In the claim ledger
^ a bRuby and Sapphire's atk01_accuracycheck, which Emerald keeps with renamed constants, scales a move's percent accuracy by one combined stage ratio, applies Compound Eyes (×1.3), Sand Veil in a sandstorm and Hustle on physical moves (×0.8 each), and a target's evasion item, and misses when a roll from 1 to 100 exceeds the result.Codepret/pokeruby · pret/pokeemerald
How we know
```
1272        calc = gAccuracyStageRatios[buff].dividend * moveAcc;
1273        calc /= gAccuracyStageRatios[buff].divisor;
⋮
1276            calc = (calc * 130) / 100; // 1.3 compound eyes boost
⋮
1278            calc = (calc * 80) / 100; // 1.2 sand veil loss;
⋮
1280            calc = (calc * 80) / 100; // 1.2 hustle loss;
⋮
1296            calc = (calc * (100 - quality)) / 100;
⋮
1299        if ((Random() % 100 + 1) > calc)
```
src/battle_script_commands.c lines 1272–1299 at 5784633 · upstream ↗
src/battle_script_commands.c#L1099-L1252 · upstream ↗
Checked against the evidence by its author, an AI model, 2026-10-21.
Checked by a second AI model, 2026-10-21.
In the claim ledger

^ a b c d e f gCmd_accuracycheck combines the attacker's accuracy stage and the target's evasion stage into one stage, multiplies the move's accuracy in percent by that stage's ratio, and misses when a roll from 1 to 100 exceeds the result, so a 100% move at even stages never misses.Codepret/pokeemerald

How we know

1128        if (gBattleMons[gBattlerTarget].status2 & STATUS2_FORESIGHT)
⋮
1136            buff = acc + DEFAULT_STAT_STAGE - gBattleMons[gBattlerTarget].statStages[STAT_EVASION];
⋮
1149        calc = sAccuracyStageRatios[buff].dividend * moveAcc;
1150        calc /= sAccuracyStageRatios[buff].divisor;
⋮
1175        // final calculation
1176        if ((Random() % 100 + 1) > calc)

src/battle_script_commands.c lines 1128–1176 at 731ad5b · upstream ↗

588static const struct StatFractions sAccuracyStageRatios[] =
⋮
590    { 33, 100}, // -6
⋮
596    {  1,   1}, //  0
⋮
602    {  3,   1}, // +6

src/battle_script_commands.c lines 588–603 at 731ad5b · upstream ↗

Checked against the evidence by its author, an AI model, 2026-10-20.

Checked by a second AI model, 2026-10-21.

#How it works

#Implementations

#Dragon Warrior

#Final Fantasy

#Final Fantasy II to VI

#Pokémon Red and Blue

#Pokémon Crystal

#Pokémon Emerald

#EarthBound Beginnings and EarthBound

#Fire Emblem: The Sacred Stones

#Variants and combinations

#History

Related

Combines with

See also

Mentioned in

Lineage

Games

References

Sources

Code

Primary

Secondary