Head-to-head in tennis: does the H2H really predict the winner?

“He leads their head-to-head 3-0.” It is often the first number quoted before a match, and one of the most used to make a prediction. The idea is appealing: some players cause problems for others, with their style, their serve or their backhand, and a lopsided record should keep repeating itself.

We wanted to check. Across 38,123 ATP and WTA singles matches played between January 2022 and September 2026, we replayed the head-to-head record as it stood before each match, then looked at who won. The verdict is nuanced: the head-to-head does carry information, but much less than people think, and not in the way it is usually used.

Most matches have no head-to-head at all

First, an often forgotten fact: in 62% of matches, the two players had never met in our archive, which goes back to 2020. And when they had, it was usually just once: six times out of ten, the “record” comes down to a single match, typically played about ten months earlier.

• At least one previous meeting: 37.6% of matches.
• At least three meetings: 6.2% of matches.
• At least five meetings: 1.3% of matches.

The long rivalries that make headlines are the exception. A 1-0 or 2-1 record weighs very little: it is far too small a sample to tell a real matchup problem from an off day.

The head-to-head leader wins less often than the favourite

Take every match where one of the two players led the head-to-head. Who should you have followed: the leader, or the higher-rated player according to Elo, which measures overall level from who beat whom?

On the same matches, the Elo favourite wins 64.2% of the time, the head-to-head leader 57.8%. A gap of more than six points: picking the player who “usually beats the other” rather than the stronger player right now is the worse strategy. Even with a lead of three wins or more, the head-to-head only matches Elo.

Why?

Because a record tells the past of two players who have changed since. The youngster beaten three times at 19 is not the same player at 22. Elo is updated after every match against every opponent: it tracks a player’s rise or decline continuously, while the head-to-head stays frozen on a handful of dates.

But a head-to-head against the favourite does matter

That does not make the head-to-head useless. The right question is not “should it replace Elo?” but “does it add anything on top of Elo?”. To answer it, we compared matches where the favourite had the same Elo win probability, depending on whether the head-to-head was in their favour, against them, or level.

The effect is asymmetric. When the head-to-head agrees with Elo, it adds almost nothing: the favourite wins one point more than expected, too small a gap to tell apart from chance over this many matches. When it contradicts Elo, however, the favourite loses 2.6 points of probability, a clear gap (a one-in-five-thousand chance of showing up by luck).

It makes sense: if the favourite already dominates their opponent, their head-to-head lead only confirms what Elo knew. If they have lost to that opponent several times despite a higher level, it signals a matchup problem Elo cannot see.

The “bogey player” is real, but rare

The effect grows with the lower-rated player’s lead. One win ahead costs the favourite about 2 points; two wins, about 3 points; three or more, and the favourite wins only 50.6% of matches against 59.6% expected: close to a coin flip. That is the true bogey player. But it concerns just 162 matches out of 38,123, fewer than one match in two hundred, and the figure rests on too few cases to be taken to the decimal.

An old defeat weighs more than a recent one

The last surprise is the most counter-intuitive. You might think the latest meeting counts most. The player who won it does win the next one 56.7% of the time. But at equal Elo probability, a recent defeat of the favourite changes almost nothing, while an older one does.

The explanation lies in Elo itself. A recent defeat has already lowered the favourite’s rating: it is built into their probability, so the head-to-head counts it twice. A defeat from two years ago has faded from Elo over the following matches, while the matchup problem it revealed may still be there. That is what the head-to-head really adds: a long memory of the encounters, which the rating forgets.

How to read a head-to-head before a match

• Start with current level. A favourable head-to-head does not make a weaker player the favourite.
• Ignore records of one or two matches. 1-0 or 2-1 is mostly noise.
• Take seriously a favourite who keeps losing to this opponent. It is the only case where the head-to-head changes the picture, especially with three defeats or more.
• Don’t overrate the last meeting. If it is recent, the rating already accounts for it.
• Check the surface of past meetings. A 3-0 built on clay may say little about a match on grass (see what the surface really changes).

That is exactly how the Pronocast algorithm works: the head-to-head is one input among others, weighted by the age of each meeting, and the model learns how much it should count against Elo, form and market odds. You can see the result in today’s predictions.

Method

ATP and WTA singles matches from January 1, 2022 to September 24, 2026 where both players had already played at least twenty matches: 38,123 matches. The head-to-head is the raw record of meetings between the two players before the match, across all tours (Challenger and ITF included), since the start of our archive in January 2020: older meetings are not included. The favourite is the higher-rated player on the global Elo computed by Pronocast before the match. To compare at equal probability, each match is set against matches with no head-to-head leader whose favourite’s Elo probability falls in the same 2.5-point band. The −2.6-point gap is significant (p = 0.0002); the +1.0-point gap is not (p = 0.06).