Prop LabSpeed, faceoffs and lines
NHL Faceoff Matchups: Ignore the Head-to-Head, Watch the Hands
A center’s record against one opponent feels like inside information. In our test of 73,496 draws, it made predictions worse. What really decides a matchup is which hand each center shoots with and which dot the puck drops on.
By Edgehalla ResearchPublished 6 min read
The head-to-head trap
Faceoff head-to-head records are easy to find, and they look decisive. Crosby went 64–32 against Suzuki (66.7%) in our data. Surely that tells you who wins tonight?
Mostly, it does not. Our model puts Crosby at about 55–57% against Suzuki at every dot. The rest of that 66.7% is luck. Even the biggest rivalries are small samples. Over three seasons, only 213 pairs of centers met 50 or more times, and only 9 met 100 or more. The most frequent pair, Staal vs Hischier, met 133 times.
We tested this properly. We trained on 2023-24 and 2024-25 and predicted every 2025-26 draw where the two centers had already met at least 10 times (24,877 draws). Then we compared raw head-to-head records with a skill model and with blends of the two.
Does the head-to-head record help predict the next draw?
Log-loss on 2025-26 draws between centers with 10+ prior meetings (lower is better)
Raw head-to-head records predicted worse than a coin flip; trusting them less and less improved the score until it matched the skill model.
The best blend gives the head-to-head record almost no weight, and it still only ties the model. After controlling for skill, home, zone, dot side and strength, matchup-specific edges have a standard deviation of only about 1–2 points.
What really matters: hands and dots
A center’s strong side is the dot where his forehand faces the middle of the ice. For a left shot, that is the left dot in his attacking direction. On the strong side he can pull the puck back on his forehand. On the weak side he is fighting his backhand.
Faceoff win rate by situation
Share of draws won
Strong-side centers beat weak-side centers 56.7 percent of the time; home ice at center is a coin flip.
This creates real, predictable matchups. When two opposite-handed centers meet, they are always both strong or both weak, so the dot does not matter. When two same-handed centers meet, one of them is always on his weak side, and the edge flips with the dot.
Staal and Hischier both shoot left. Our model gives Staal 57.9% on one offensive-zone dot and 44.1% on the other: a swing of about 14 points from the same two players.
Coaches already know this. Left-shot centers take draws on their strong side 62% of the time, and right-shots 76% of the time. About two-thirds of all draws (67.5%) are taken by left shots.
What does not matter much
- Home ice at center. The rulebook makes the visitor put his stick down first at center ice. Home teams still win only 49.9% there. Across all dots, home ice is worth about a point.
- Zone. A center in his offensive zone vs one in his defensive zone wins 50.15%. Zone barely matters on its own.
- The arena. Faceoffs show no scorer bias by building. Each arena’s home faceoff residual is about the size of pure noise, and it does not repeat year to year (r −0.01 and −0.13).
One more real effect is about fatigue, not skill. After its own icing, a team cannot change lines and wins only 43.9% of the next defensive-zone draws, against a usual 52.7%. More on that in game script and momentum myths.
Faceoff skill is real, and stable
Unlike head-to-head records, overall faceoff skill repeats. A center’s win rate in one half of his draws predicts the other half at r 0.76 (about 417 draws per half). After removing noise, true talent has a standard deviation of about 4.8 points.
That is why our model, a Bradley–Terry rating with home, zone, side and strength terms, scored 0.67974 log-loss on the untouched 2025-26 season. Raw faceoff percentages scored 0.68585 and a coin flip 0.69315. The gains look tiny, but on 73,496 draws they are clear and consistent.
Faceoffs won props and fantasy
Pick’em sites post faceoffs-won lines for top centers, and many fantasy leagues count faceoffs won (FOW). So does the model help there?
Projecting faceoffs won, 2025-26 backtest
The faceoff model tied a simple baseline on faceoffs won per game, because draw counts drive most of the error.
Honestly, it is a tie. Our faceoffs-won projection missed by 1.568 per game on average. A simple baseline missed by 1.564. The win rate is not the hard part. The number of draws is: it swings with penalties, icings and score, and our draw-count error was about 2.5 per game.
The matchup still moves the needle at the edges. For a top center, the expected win rate runs from about 51% to 58% depending on the opponent and dots, worth roughly 1.2 faceoffs won a game. Only 6 centers averaged 10+ FOW per game in 2025-26, and 31 averaged 8+, so a one-win swing matters for a prop line.
One thing faceoffs are not: a route to goals. Winning an offensive-zone draw instead of losing it is worth only about 0.013 expected goals at 5-on-5. Faceoffs matter as a fantasy category, not as a source of points.
- Ignore raw head-to-head records, or shrink them very hard toward the skill rating.
- Check hands and expected dots. Same-handed matchups carry the biggest swings.
- For props, project the draw count first. Tonight’s numbers are on faceoffs tonight.
Frequently asked questions
Do faceoff head-to-head records predict who wins the draw?
Barely. On 24,877 test draws between centers who had met 10+ times, raw head-to-head records scored a log-loss of 0.718, worse than a coin flip at 0.693. A skill model with handedness and dot side scored 0.684.
What is a strong-side faceoff?
It is a draw on the dot where the center’s forehand faces the middle of the ice: the left dot for a left shot, the right dot for a right shot, in his attacking direction. A strong-side center beats a weak-side center 56.7% of the time (n 93,000).
Do home teams win more faceoffs in the NHL?
Only slightly overall, about a point. At center ice, where the visitor must put his stick down first, the home team still won just 49.9% of draws in 2025-26.
Why are faceoffs-won projections so hard?
Because the number of draws a center takes is unpredictable. Our 2025-26 backtest missed draw counts by about 2.5 per game, so even a better win-rate model tied a simple baseline on faceoffs won (1.568 vs 1.564 average error).
Who has the best faceoff record against a specific center?
Raw records exist for every pair, but most are small samples: over three seasons, only 213 pairs met 50+ times. Our faceoff pages show each matchup by dot, with the raw record shrunk toward the model so luck does not dominate.
More in Speed, faceoffs and lines
Sources and method
- NHL stats (NHL.com)
- NHL API reference (community documentation)
- Bradley–Terry model
- Cross-entropy (log-loss)
- Split-half reliability and the Spearman–Brown formula
- Responsible Gambling Council (Canada)
- ConnexOntario (1-866-531-2600)
- National Council on Problem Gambling (1-800-GAMBLER)
Research and analysis, not betting advice. Past results do not guarantee future returns; bet only what you can afford to lose.