OnlineBlackjack Tips
OnlineBlackjack Tips

Guides

18+An educational guide for adults. It has no casino links and no offers. If gambling stops being fun, get help.

Home / Betting systems

The Kelly criterion: sizing bets when you have an edge

Betting systems that change stakes after wins and losses cannot create an edge. The Kelly criterion is different: it is a rule for how much to stake when you really do have one. Understanding it also explains why, at normal tables, it says to bet nothing.

Updated November 2026 · 6 minutes · For adults, 18+

A wooden slide rule lying on a dark table beside a single ace of spades
Illustration generated with AI. A wooden slide rule lying on a dark table beside a single ace of spades.

In brief

For an even-money bet with win probability p, Kelly says to stake 2p − 1 of your bankroll. For blackjack the fraction is roughly your edge divided by the variance of one hand (about 1.3). If your edge is negative, the answer is zero.

Where it comes from

In 1956, John L. Kelly Jr. of Bell Labs published "A New Interpretation of Information Rate". It showed how to size repeated bets to maximise the long-run growth rate of a bankroll. Edward O. Thorp, who wrote about card counting, later applied the approach to blackjack and to investing.

The idea

If you know you have an edge, betting too little wastes it, and betting too much risks ruin. Betting a fixed fraction of your bankroll each time, with a size depending on the edge, grows the bankroll fastest in the long run.

For a simple even-money bet that wins with probability p and loses with probability 1 − p, the fraction is:

f = 2p − 1

Examples:

Win probability Edge Kelly fraction of bankroll
50% 0 0%
51% 2% 2%
52% 4% 4%
55% 10% 10%
46.3% −7.4% Bet nothing

The last row is roughly an ordinary blackjack hand in the simplified model from the Martingale guide, where you win a little less than half your decided hands. A negative result means that Kelly says not to bet at all.

Kelly and blackjack

Blackjack is not a simple even-money bet: it has blackjacks, doubles, splits and pushes. A common approximation is that the Kelly fraction equals your edge divided by the variance of a hand. One hand of blackjack has a standard deviation of about 1.14 bets, so a variance of about 1.3.

A counter with a 1% edge would therefore stake roughly 0.01 / 1.3 ≈ 0.8% of the bankroll on a hand. With an edge of 0.5% it is about 0.4%. That is a small bet relative to the bankroll, which shows why successful counters need large bankrolls.

The catch for ordinary players

At a standard online table, the house has an edge, typically between a few tenths of a percent and about 1% with good rules and basic strategy (see house edge explained). With a negative edge, the Kelly fraction is negative, meaning no bet. The formula gives you the answer to "how much should I stake to grow my money": none.

That is not a reason you cannot play for entertainment. It is a reason not to think of blackjack as an investment, or to treat any staking pattern as a strategy for making money. Entertainment has a cost, and the sensible approach is to decide how much you are willing to pay for it.

Fractional Kelly

Even for people with a real edge, full Kelly produces large swings. Many use half Kelly or less because the edge is uncertain and the full bet is volatile. This is a risk-management choice, not a different system.

Takeaways

  • Betting systems after wins and losses (see Martingale and positive progressions) cannot make an unfavourable game favourable.
  • Kelly sizing helps only with a genuine advantage.
  • With a negative edge, the best "bet size" for money-making purposes is zero. Treat any play as paid entertainment from a fixed budget.

Quick answers

Can I use Kelly to bet at an online table?

It tells you how to size a bet with an edge. At a table with a house edge the answer is zero.

Does Kelly guarantee growth?

No. It maximises the long-run growth rate if your edge estimate is right, and swings can be large.

Why half Kelly?

It cuts volatility and protects against an overestimated edge, at a small cost in growth.