Clay, hard or grass: what the surface really changes in tennis

Nadal on clay, Federer on grass: tennis has always had its specialists, and the idea that a match is decided first by the surface is one of the most common beliefs among people who make predictions. It is partly true. But once you measure it, it holds two surprises: a player’s overall level predicts better than their record on the day’s surface, and the surface where favourites fail most often is not the one most people expect.

We went back over 36,210 ATP and WTA singles matches played between January 2022 and September 2026, between players who already had at least twenty matches behind them. For each one, we looked at who was the favourite before the first point, according to two measures, and who won. Here is what the surface really changes, and what it doesn’t.

Four surfaces rather than three

People talk about three surfaces, but models often separate four, because indoor hard courts don’t play like outdoor ones.

• Clay slows the ball down and makes it bounce high. Rallies are longer, the serve earns fewer free points, and fitness matters more. It is the surface of Roland-Garros and of the whole European spring season.
• Grass is the opposite: fast ball, low and sometimes irregular bounce, a clear edge to the server. Its season lasts only about five weeks, around Wimbledon.
• Hard courts sit in between and are by far the most common: two Grand Slams, most Masters 1000 events and the bulk of the calendar.
• Indoor hard courts, with no wind or sun, favour big servers and steady ball strikers. They host the end of the season, up to the Finals.

First surprise: overall Elo predicts better than surface Elo

The basic tool of most models is the Elo rating, which measures a player’s strength from who they have beaten. You can keep a single one, over all matches, or one per surface, computed only from matches played on that surface. Intuition says the second should predict a clay or grass match better. The numbers say the opposite.

On every surface, the favourite picked by overall Elo wins more often than the one picked by surface Elo. The gap is small on hard courts, 2.2 points on clay, and it becomes huge on grass: more than five points.

The reason: sample size

An Elo rating is only reliable with plenty of matches. Yet at the time of a match in our sample, a typical player had already played 90 matches on hard courts, 57 on clay, and just 12 on grass (median values). Twelve matches is barely two or three grass seasons: too few to tell a player’s real level from a good or bad week. Overall Elo, fed by every match, remains the best estimate of a player’s strength, including on a surface where they rarely play.

When the two measures disagree

In 21% of matches, the two Elo ratings don’t pick the same favourite: the player rated higher overall is rated lower on the day’s surface. That is the classic “specialist” against a player who is stronger on paper. Who wins then?

The stronger player overall wins more than half the time, everywhere. On hard courts it is close to a coin flip: both measures carry comparable information. On grass, however, the so-called specialist wins only 42% of these matches. A good past grass season weighs less than the player’s current level.

What a model does with this

It doesn’t pick one. Simply averaging the two probabilities gives 65.0% winning favourites overall, barely better than overall Elo alone, but 65.4% on hard courts, outdoors and indoors, where the two measures add up usefully. On grass, the same average costs a point and a half. That is exactly what a tennis prediction algorithm learns to do: read both values and weigh how much to trust each one depending on the surface, instead of treating them as a single number.

Grass, the least predictable surface

Second surprise, partly a consequence of the first: for the same gap in level, the favourite wins less often on grass than anywhere else. The table compares, for matches where overall Elo gave the favourite the same probability, the share actually won.

Between two close players, a grass match is nearly a lottery: the slight favourite wins only 51.1% of the time, against 55.3% on clay. The most common explanation lies in the game itself. On a surface where the server almost always holds, a set is decided by one or two break points, or a tiebreak, and a handful of points is enough to overturn the hierarchy. Add a very short season and unreliable surface ratings, and differences in level become harder to estimate.

Clay, the most readable

At the other end, clay is the surface where favourites deliver most. Long rallies multiply the number of points played, and the more points a match has, the less a bad patch or a lucky shot weighs on the result. The stronger player has time to impose their level.

One thing every surface shares

In every row of the table, the favourite wins slightly less often than its Elo probability says: a player given 75% actually wins around 71%. Raw Elo is a little overconfident. That is why a serious model never publishes the Elo probability as is, but a probability recalibrated on real results, and compares it with market odds when they exist.

What to take away when reading a prediction

• Surface matters, but less than overall level. A stronger player stays the favourite, even on a surface they like less.
• Be wary of short surface records. A player who is “great on grass” over fifteen matches may just have had one good season.
• On grass, widen your margin of error. A 60% favourite loses there more often than elsewhere, and Wimbledon upsets are not accidents.
• On clay, favourites hold up better. Differences in level translate most faithfully into wins there.

Pronocast’s daily predictions show the surface of every match, and each player’s page gives their record by surface, with a minimum number of matches before concluding they are better on one than another.

Method

ATP and WTA singles matches from 1 January 2022 to 24 September 2026, with a known surface and where both players had already played at least twenty matches: 36,210 matches in total. Indoor hard courts are separated from outdoor ones. The Elo ratings are those computed by Pronocast before each match: no information from after the first point enters the figures. A favourite’s probability comes from the standard Elo formula, without recalibration, to compare surfaces with the same tool. The model published on the site combines these ratings with other variables and with market odds.