Tennis as Game Theory: Mixed Strategies on Serve

Most serve coaching focuses on mechanics — toss, drive, pronation, contact. That’s necessary. But there is a layer above mechanics that almost no club coach addresses, and it is where the math says points are won and lost at the top of the game. That layer is strategic, not technical. It is game theory.

The serve is a strategic choice as much as a physical action. The server decides — every single point — where to aim, what spin to use, and how hard to hit. The returner must guess, or read, or both. The interaction between server and returner is a repeated game with mixed strategies, and the principles of game theory predict surprisingly precise things about how the best players actually serve.

This article is the case for thinking about your serve choices as a strategy, not just a stroke.

The Basic Setup: A 2×2 Serve Game

Strip the situation down to its simplest form. The server can aim at one of two targets — call them T (down the middle) and Wide. The returner can lean toward one of two anticipation directions — also T or Wide. Each of the four combinations produces a different win probability for the server:

• Server T, Returner anticipates T → server wins, say, 50% of points.
• Server T, Returner anticipates Wide → server wins, say, 75% of points.
• Server Wide, Returner anticipates T → server wins, say, 80% of points.
• Server Wide, Returner anticipates Wide → server wins, say, 45% of points.

These numbers are illustrative, not measured. But the structure is what matters. Each server target has a “good” outcome (caught the returner leaning wrong) and a “bad” outcome (returner anticipated correctly). The server’s job is not to find the single best target — that doesn’t exist as a stable strategy — but to find the right mix of targets that maximizes long-run win rate.

This is the core game-theory insight: in a competitive interaction where the opponent adapts, the best strategy is usually a mixed strategy, not a pure one.

What Goes Wrong With Pure Strategies

A server who serves only down the T is playing a pure strategy. The first three points might work — the returner doesn’t know the pattern. By the fourth point, the returner is camping on the T side, and the win rate collapses.

The same is true of any pure strategy. Serve only wide, only to the body, only with kick spin — any consistent pattern can be exploited. The returner’s job is to detect the pattern. The server’s job is to make detection impossible by mixing unpredictably.

This is why “I have a great wide serve, I’ll just use it every time” doesn’t work past about the third use. The serve degrades in effectiveness not because the serve got worse but because the returner learned what to expect.

What the Math Says About the Right Mix

Game theory’s solution concept here is the Nash equilibrium — the mixed strategy where neither player can improve their results by changing their own probabilities, given the opponent’s probabilities.

In the simplest 2×2 example above, the math says the server should serve T about 40% of the time and Wide about 60% of the time. The exact numbers depend on the win-rate matrix, but the structural prediction is: the server should mix in proportion that equalizes the returner’s payoff between leaning each direction. If the returner has no advantage from anticipating either way, they cannot exploit the server.

Studies of actual ATP service patterns find that top servers approximate Nash equilibrium probabilities quite well (Walker & Wooders, 2001). When researchers analyzed hundreds of professional serves and computed the implied optimal mix, the actual mix used by top players matched within a few percentage points. Players who deviate significantly from optimal mixes — too much wide, too much T — produce slightly lower win rates than they could.

This is one of the few sports-strategy areas where game theory makes specific, testable predictions and the data validates them.

Why Amateur Servers Get the Mix Wrong

In coaching practice, the most common amateur error is to over-serve to the favorite spot. A player with a good wide serve will hit wide on 70%+ of first serves. Within a few games, the opponent is leaning hard toward the wide side, and the server is producing weak points on every wide serve while neglecting their second-best target.

The correction isn’t “use your wide serve less.” It is “use it in a proportion that makes the opponent uncertain.” Practically, that usually means cutting the favorite target to 50–55% of serves, with the remaining 45–50% distributed among the other targets in ways the opponent can’t easily predict.

This is harder than it sounds. Humans are bad at randomness. We pattern even when we try not to. The fix is to use external randomness — a pre-decided rotation, a coin flip in your head, a count of opponent reactions — to enforce variability that your own brain wouldn’t produce.

Reading Patterns Is the Returner’s Half of the Game

The returner has a corresponding strategic problem: how much to anticipate, and what to anticipate.

A returner who never anticipates returns purely on reaction. They play a “neutral” strategy that gives away no information but also doesn’t exploit any pattern in the server. Their return win rate is the average of all the server’s targets.

A returner who anticipates correctly raises their win rate against that target — but loses badly when the server mixes against them. They have to be right more often than wrong.

The middle ground is what good returners do: anticipate selectively, on cues that justify the bet. Toss location, shoulder lean, pre-serve body language — these are real information leaks. A returner who reads them well can shift their anticipation 10–20% of the time and gain a measurable edge. A returner who anticipates on no information loses to the server’s mix.

Game Theory and the Second Serve

The second serve is where game theory gets especially interesting. The cost of a missed serve is much higher (double fault), so the server’s risk-reward calculation changes. They cannot use the same mix of placement and pace as on first serves.

The honest implication is that second serves are inherently more constrained, more predictable, and more attackable. The returner can lean further forward, can anticipate the kick, can plan an aggressive return. The game-theory situation tilts in the returner’s favor.

This is why second-serve return-points-won is the single most predictive statistic of match outcomes at the elite level. The structural game-theory advantage that the server has on first serves is reduced or reversed on second serves. Players who can attack second serves — who have read the constraints and built a return mindset around them — accumulate the advantage over a match.

When Pure Strategies Actually Work

A footnote worth including: there is one situation where a pure strategy is correct, and it’s when the gap between best and worst targets is so large that no mix can exploit the opponent’s adaptation.

If you have a 200 km/h flat serve to the T against an opponent who returns balls at 130 km/h max, the math says serve T almost every time. The opponent can’t physically catch up to the wide serve advantage you’d gain by mixing — the T serve is just so much better in this matchup that mixing costs you more than it gains.

This is rare. Most players don’t have a target that is 30+ percentage points better than their second-best. For everyone else, the math says mix.

How to Build Mixed Strategies in Training

Three drills I use:

Drill 1: Pre-decided serve sheet. Before a practice match, the server writes down their target sequence for the first ten games — generated by external randomness (dice, coin, rotation). They serve from the sheet. The point is to feel what it’s like to serve a random mix rather than relying on instinct.

Drill 2: Returner-feedback drill. Two players, one returner reports out loud what target they anticipated on each point. After 20 points, the server sees their own pattern through the returner’s eyes. Patterns the server didn’t know they had become visible.

Drill 3: Counter-tendency drill. The server identifies their two most common serves and is required to use their third serve target on at least 30% of points for a full set. This breaks pure-strategy reliance and builds the secondary serves that a real mix requires.

What This Doesn’t Mean

A few clarifications worth flagging.

It doesn’t mean serve placement is random. The mix is conditional — different mixes for different score situations, different opponents, different serve speeds. Within a context, the mix should be unpredictable. Across contexts, it can shift.

It doesn’t replace technique. A serve that mixes well but is mechanically poor will lose. Game theory is the layer above technique, not a substitute for it.

It doesn’t apply only to the serve. Cross-court vs down-the-line decisions, approach shot direction, lob vs pass — all have game-theory structure. The serve is just the cleanest example because the choice is purely the server’s.

One Thing to Do on Court Tomorrow

Before your next practice match, write down a sequence of ten serve targets, generated by flipping a coin. Five “T” and five “Wide” in whatever order the coin gives. Serve the sequence in order, regardless of how the previous point went. After ten serves, write down what you noticed — about your own discomfort with the unfamiliar choices, about your opponent’s anticipation patterns, about the points that worked when you served somewhere you wouldn’t have chosen.

Most players, doing this for the first time, are surprised by two things. The unfamiliar choices feel awful — and they win at the same rate as the familiar ones, because the unfamiliarity is the point. Mixed strategies feel wrong precisely because they are not pattern-recognizable. That feeling, after enough exposure, is the feeling of doing it right.

About the author: Emre Köse is a tennis coach at Beykoz Tenis Kulübü in Istanbul, with 12+ years on court. He holds a BSc in Coaching Education from Marmara University, Faculty of Sport Sciences.

Related in this series: Reading patterns · Building a game plan · Risk management

Selected reading:

• Walker, M., & Wooders, J. (2001). Minimax play at Wimbledon. American Economic Review.
• Klaassen, F. J. G. M., & Magnus, J. R. (2014). Analyzing Wimbledon: The Power of Statistics. Oxford University Press.
• Hsu, S. H., Huang, C. Y., & Tang, C. T. (2007). Minimax play at Wimbledon: comment. American Economic Review.