Across 14,082 ATP and WTA matches, the player who landed more of their first serves won 55.9% of the time, barely better than a coin flip. The player who won more points behind their first serve won 84.9% of the time. Broadcasts show the first number; the second is the one that predicts the result.
Our ingestion pipeline used to drop the statistics block the feed sends with every finished match. After backfilling 2026 we had 14,082 ATP and WTA matches with complete serve numbers for both players, and we tested the serve stat broadcasts quote most: first-serve percentage.
The test is within a match, same court and conditions: did the player with the higher number win?
| The player who had the higher… | won the match |
|---|---|
| First-serve percentage (serves landed) | 55.9% |
| First-serve points won | 84.9% |
Landing more first serves wins 55.9% of the time. Winning more of the points when the first serve lands wins 84.9%.
Sorted by how far apart the two players were on each stat:
| Gap between the players | Higher first-serve % wins | Higher first-serve points won wins |
|---|---|---|
| 0 to 2 points | 49.6% | 53.5% |
| 2 to 5 points | 52.7% | 65.9% |
| 5 to 10 points | 54.4% | 82.4% |
| 10 to 20 points | 59.4% | 95.0% |
| Over 20 points | 66.9% | 99.3% |
Out-landing the opponent by more than twenty points still loses one match in three. A comparable gap in first-serve points won decides the match 99.3% of the time.
The player controls it, at a cost: taking pace off raises the percentage. Grouped by first serves landed, the value of each landed serve falls:
| First serves landed | WTA: first-serve points won | ATP: first-serve points won |
|---|---|---|
| Under 55% | 64.6% | 69.8% |
| 55 to 60% | 63.9% | 69.7% |
| 60 to 65% | 63.0% | 69.6% |
| 65 to 70% | 61.5% | 68.9% |
| Over 70% | 60.2% | 65.4% |
From the bottom group to the top, effectiveness drops five points in the WTA and four and a half in the ATP.
Below 55%:
| First serves landed | WTA: FS points won | WTA: match win | ATP: FS points won | ATP: match win |
|---|---|---|---|---|
| 50 to 55% | 64.9% | 46.3% | 69.7% | 44.1% |
| 45 to 50% | 63.5% | 36.5% | 69.8% | 41.5% |
| 40 to 45% | 65.6% | 33.3% | 70.2% | 39.7% |
| Under 40% | 65.4% | 31.0% | 71.1% | 41.4% |
Effectiveness gains about a point on the way down while the match-win rate falls sharply: land under 40% of first serves and WTA players win 31% of their matches. Every missed first serve starts the point on a second serve, which gets attacked. The smallest bucket holds 29 WTA player-matches, so the exact figures are rough, but the direction is clear: no hidden value below 55%, and losers at both extremes.
The trade-off is well known to statisticians. Whether to hit two big first serves instead of a big one and a safe one has been a published optimization problem since 1971 [1], and George showed two years later that there is a sizeable region where two big serves is better [2]. Klaassen and Magnus measured how far professionals sit from their own optimal serve strategy at Wimbledon: about 1.1% of a service point for men, 2.0% for women [3]. With most players near their optimum, differences in first-serve percentage mostly reflect style, not quality.
Players in the over-70% group do win more matches: 57.1% in the WTA and 57.4% in the ATP, against about 43% for the under-55% group. That comparison is across different players, and good servers both land more first serves and win more matches. The within-match 55.9% is the fair test. It is a real effect over 14,082 matches, just small: six extra correct calls per hundred matches, against thirty-five for first-serve points won.
We expected second serve to separate the tours. It does not: winning more second-serve points takes the match 77.9% of the time in the WTA and 77.1% in the ATP.
The first version of this article claimed service points decide 93.7% of ATP matches but only 78.3% of WTA matches, and built a section on the return game on it. Recomputed from raw point counts the figures are 92.2% ATP, 94.7% WTA. We could not reproduce the old WTA figure; it was a computation error on our side, and the section is removed.
First-serve percentage quoted as evidence is evidence of very little. First-serve points won is the serve number that predicts. Our model prices matches from the points-won side and does not use first-serve percentage.
Next: the feed carries set-by-set serve rows, which would show whether players serve safer right after being broken and what that costs them.
Method: 14,082 ATP and WTA singles matches from 2026 with complete serve statistics for both players, 28,164 player-match records. Every winner-versus-loser comparison is made within a match, so opponent quality, surface and conditions are held constant. Matches tied exactly on the stat being compared are excluded from that comparison, 3.7% of the sample on first-serve percentage. Percentages are computed from raw point counts, never averaged from rounded per-match percentages. Statistics coverage begins in 2026. Correction 2026-08-07: an earlier version claimed service points decide 93.7% of ATP matches but only 78.3% of WTA matches; recomputation from raw counts gives 92.2% and 94.7%, and the tour-difference section built on the wrong figure has been removed.
References:
[1] D. Gale, "Optimal strategy for serving in tennis," Mathematics Magazine, vol. 44, no. 4, pp. 197–199, 1971.
[2] S. L. George, "Optimal strategy in tennis: A simple probabilistic model," Journal of the Royal Statistical Society Series C (Applied Statistics), vol. 22, no. 1, pp. 97–104, 1973.
[3] F. Klaassen and J. R. Magnus, "The efficiency of top agents: An analysis through service strategy in tennis," Journal of Econometrics, vol. 148, no. 1, pp. 72–85, 2009.
It helps a little. Comparing the two players inside the same match, the one who landed a higher share of first serves won 55.9% of the time. The problem is what it costs: players who landed over 70% of first serves won only 60.2% of those points in the WTA, against 64.6% for players landing under 55%, and the ATP shows the same slide, 65.4% against 69.8%. You buy consistency with effectiveness, so the number rises while the serve gets weaker. That is why it barely predicts anything.
No, and we expected the opposite. The player who won more second-serve points took the match 77.9% of the time in the WTA and 77.1% in the ATP. Service points overall are just as symmetric: winning that battle takes the match 94.7% of the time in the WTA and 92.2% in the ATP. An earlier version of this article claimed a large tour split there; that was a computation error on our side and is corrected in the method note.
Our match database goes back further, but per-match serve splits only arrive in the feed from 2026 onward. Everything in this article is restricted to matches where both players have complete first and second serve statistics, which is 14,082 matches. We would rather report a smaller honest sample than pad it with matches where the serve numbers are missing.
See today's picks: published before the match, graded in public →