Poker Variance & Confidence Calculator
Estimates the sample size required to confirm your poker win rate at a chosen statistical confidence level. Ideal for serious players questioning whether their recent results reflect true skill or short-term variance.
Last updated: May 2026
Compare with similar
About this calculator
Poker results follow an approximately normal distribution over large samples, allowing the use of standard statistical confidence intervals. The number of hands needed to confirm a win rate within a target precision (±X BB/100) at a given confidence level is N = (Z × standardDeviation / targetPrecision)², where Z is the Z-score for the chosen confidence level (1.645 for 90%, 1.96 for 95%, 2.576 for 99%). Variables: handsPlayed (current sample), actualWinRate (observed win rate in BB/100), standardDeviation (your tracked SD in BB/100, typically 80–120 for cash, 150–200 for tournaments), confidenceLevel (Z-score), targetPrecision (acceptable margin in BB/100). The result tells you how many hands you must play to make your observed win rate statistically reliable at that confidence and precision. For example, proving a 5 BB/100 win rate to within ±2 BB/100 at 99% confidence requires roughly 1.6 million hands. Edge cases: the formula assumes your true win rate is stationary, which is rarely true as you change games, stakes, or playing style. It also assumes hand outcomes are independent — table dynamics, tilt, and selection effects violate this assumption. Sample-size requirements scale with the square of standard deviation, so tournament players need vastly more events than cash-game players for the same confidence. Below 10,000 hands the central-limit assumption breaks down and you should use bootstrap methods instead.
How to use
Example 1: Determine hands needed to confirm your win rate to ±2 BB/100 at 99% confidence with SD = 100 BB/100. Step 1: Z = 2.576 (99% confidence). Step 2: N = (2.576 × 100 / 2)² = (128.8)² ≈ 16,590. This is in units of 100 hands (since SD is per 100), so total = 16,590 × 100 = 1,659,000 hands. Verify: the rule of thumb 'you need a million hands to know your win rate' holds — tight precision and high confidence require enormous samples. Example 2: Looser settings — 95% confidence, ±5 BB/100 precision, SD = 90 BB/100. Step 1: Z = 1.96. Step 2: N = (1.96 × 90 / 5)² = (35.28)² ≈ 1,245 units of 100 hands = 124,500 hands. Verify: this is closer to one year of full-time online cash-game play for a moderate player — still a long road, but achievable.
Frequently asked questions
How many hands do I need to play to prove my poker win rate is real?
At 99% confidence with SD 100 BB/100 and ±2 BB/100 precision you need roughly 1.6 million hands — more than most players accumulate in a lifetime. At 95% confidence and ±3 BB/100 precision you still need around 400,000 hands. This is why short-term results — even 50,000 to 100,000 hands — are heavily influenced by variance. Poker tracking software like PokerTracker 4 and Hold'em Manager 3 visualize your win-rate confidence interval as it narrows over time, making it easier to interpret your results honestly. Live poker players who play 30 hands per hour need decades to reach the same confidence as online grinders running 6–8 tables.
What standard deviation should I use for my poker variance calculations?
For No-Limit Texas Hold'em cash games, standard deviation typically ranges from 80 to 120 BB/100 depending on style and stakes. Loose-aggressive and high-stakes players tend to have higher SD (100–130 BB/100) due to larger pot-to-stack ratios. Tight-passive players may be closer to 70–90 BB/100. Pot Limit Omaha has much higher variance — typically 150–250 BB/100. Tournament players see 150–200 BB/100 in MTTs and even higher in flat-payout structures. Your actual SD is tracked automatically by PokerTracker or Hold'em Manager — use that figure rather than a rough estimate. Underestimating SD gives a falsely optimistic picture of your result reliability and bankroll requirements.
Why can a winning poker player show a losing record over 50,000 hands?
A player with a true win rate of 5 BB/100 and SD of 100 BB/100 has a 95% confidence interval of roughly ±8 BB/100 over 50,000 hands — meaning their observed win rate could legitimately range from −3 to +13 BB/100. This wide interval makes 50,000 hands statistically insufficient to distinguish a winner from a breakeven player. The mathematics of variance require hundreds of thousands of hands to narrow the interval to meaningful precision. This is why emotional bankroll decisions based on small samples (dropping down after a losing month, moving up after a hot streak) are so destructive to a poker player's long-term development. The same logic applies to evaluating an opponent — never assume a player is bad based on one losing session.
What are common mistakes when using variance and confidence calculations?
Treating recent results as predictive — assuming a hot 20,000-hand sample means your win rate has improved — ignores how wide confidence intervals are at small samples. Using an SD value that is too low (out of optimism rather than data) makes the sample-size requirement look smaller than it really is. Confusing 95% confidence with 95% probability of winning each session — confidence intervals describe long-run averages, not session outcomes. Applying the formula to a non-stationary win rate (e.g., when you have been moving up stakes) gives a meaningless number because your true win rate has changed during the sample. Ignoring rake when computing your observed win rate (especially at micro-stakes) overstates your edge by 4–8 BB/100 — most tracking software reports rake-adjusted win rate; use that figure. Mixing live and online hands together is also wrong because the variance profiles differ.
When should I NOT use this variance calculator?
Tournament players measuring ROI rather than BB/100 should use ROI variance formulas instead — tournament results follow a different distribution dominated by occasional large cashes. Players with fewer than 10,000 hands should not draw any conclusions from win-rate data; focus on game improvement and hand history review rather than statistical analysis. Players who multi-table 12+ tables of high-variance formats (zoom, fast-fold pools) have inflated SD that breaks the per-table-stationary assumption. Mixed-game players (HORSE, 8-Game) have variance that depends on the rotation; use game-specific SD values. Cash-out, transfer, and bonus hands that aren't true gameplay should be excluded from the sample before computing win rate. Live tournament samples below 100 events are essentially noise — use field-size-adjusted ROI confidence calculators (Sharkscope's tools) rather than per-hand variance math.