pucklogs

Home / Articles / What Is Expected Goals (xG) in Hockey? A Plain-English Guide, Plus xG vs Corsi vs Fenwick

What Is Expected Goals (xG) in Hockey? A Plain-English Guide, Plus xG vs Corsi vs Fenwick

Examples use 2026-27 NHL data as of Nov. 12, 2026. Five-on-five team numbers are from MoneyPuck, and shooting percentages are from NHL.com.

Short answer: expected goals, or xG, estimates how many goals a team or player “should” have scored given the shots they took. Each shot gets a probability of becoming a goal based on things like distance, angle and shot type. Add those probabilities up and you get expected goals. If a team creates 3.1 xG in a game, an average shooter facing an average goalie would score about three goals on those chances.

xG is now the most widely used public measure of shot quality in hockey. It doesn’t replace goals. It explains them.

How an xG model works

An xG model is trained on hundreds of thousands of past shots, each labelled goal or no goal. It learns which features make a goal more likely. For example, MoneyPuck’s model, rebuilt for 2026-27 and trained on more than 289,000 shots from 2023-24 through 2025-26, uses:

  • Shot location: distance and angle to the net
  • Shot type: wrist, slap, backhand, tip and so on
  • The previous event: what happened before the shot (a takeaway, hit or faceoff), where, and how many seconds earlier. This is how models capture rebounds and rush chances.
  • Shift length: how long the defenders have been on the ice
  • Manpower: five-on-five, power play, three-on-three
  • Off-wing shooting: the shooter’s handedness compared with which side of the ice they’re on
  • Empty net: whether the goalie was pulled

A common illustration, used in Wikipedia’s analytics primer, puts a shot from the slot at about 0.30 xG and a shot from the point at about 0.02. At five-on-five this season, most NHL teams generate between about 1.6 (Vancouver) and 2.65 (Carolina) expected goals per game, per MoneyPuck.

How accurate is it? Model builders measure this with a statistic called AUC, where 0.5 is a coin flip and 1.0 is perfect. MoneyPuck reports an overall AUC of 0.807 for its current model, up from 0.784 for the old one. In practice, that means xG is good at telling dangerous shots from harmless ones, but individual shots still carry plenty of randomness.

xG vs Corsi vs Fenwick

All three count shots. They differ in which shots they count, and whether they weigh them.

  • Corsi (the NHL calls it SAT, for shot attempts) counts every attempt: shots on goal, missed shots and blocked shots. It is the broadest and most stable measure of who has the puck.
  • Fenwick (the NHL’s USAT, or unblocked shot attempts) leaves out blocked shots. The idea is that a blocked attempt never had a chance to score, and that shot-blocking can be a defensive skill. Most xG models are built on unblocked attempts.
  • Expected goals takes the unblocked attempts and weighs each by its quality.

The usual team version of each is a percentage, such as CF% (Corsi For %), FF% or xGF%: the share of the metric a team gets while it’s on the ice. Above 50 percent means a team is outplaying its opponents on that measure.

Here’s the practical trade-off. Corsi has the most events, so it stabilizes fastest over a small sample. xG tells you more about danger but is noisier early in a season. Analysts often look at both. When they disagree, that disagreement is usually the interesting part.

When Corsi and xG disagree: 2026-27 examples

All figures below are five-on-five, from MoneyPuck, as of Nov. 12, 2026.

  • Calgary Flames: 51.2 CF%, 47.2 xGF%. The Flames get more than half the attempts but less than half the danger, which points to plenty of perimeter volume and not enough looks from the slot. Their 28 five-on-five goals are among the league’s fewest.
  • New York Rangers: 48.1 CF%, 52.0 xGF%. This is the opposite case. They’re out-attempted, but their chances are better than the ones they allow.
  • Florida Panthers: 53.8 CF%, 51.2 xGF%. They have a strong possession share, while their shot-quality edge is smaller.
  • Carolina Hurricanes: 58.6 CF%, 59.8 xGF%. When both numbers are elite, you get the best five-on-five team in the league so far. Carolina has outscored opponents 56-30 at five-on-five.

Corsi vs Fenwick in practice: Nashville has had 293 five-on-five attempts blocked against it, and Philadelphia leads the league with 298. A team that blocks a lot looks a little better by Fenwick than by Corsi. Nashville’s numbers are 45.3 CF% and 46.4 FF%.

Goals vs expected goals: finishing, or luck?

xG is also used for individual shooters, written as ixG or individual expected goals. The gap between actual goals and ixG is often called “goals above expected.”

  • Evan Bouchard (EDM): 11 goals on 3.88 ixG in all situations. He’s shooting 21.2 percent, against a career 7.7 percent on 1,137 shots per NHL.com. That’s the biggest warning sign in this list.
  • Cutter Gauthier (ANA): 16 goals on 9.57 ixG. He’s a talented shooter on a hot streak, but the gap is large.
  • Kirill Kaprizov (MIN): 17 goals on 13.33 ixG. He’s still above expected, but closer, and his chance volume is near the top of the league.
  • Connor McDavid (EDM): 11 goals on 7.76 ixG.

Elite shooters really do beat xG over long careers, because models assume an average shooter. Over 20 games, though, gaps this size are mostly variance. MoneyPuck runs a separate “shooting talent adjusted” xG to account for proven finishers. The baseline model leaves shooter identity out on purpose.

What xG can’t see

  • Passes before the shot. Public NHL play-by-play doesn’t log most passes, so a cross-seam one-timer and a shot off a dump-in from the same spot can get similar values.
  • Screens and goalie position. These are mostly invisible in the data.
  • Recording errors. Shot locations are logged by arena scorers. MoneyPuck’s own documentation gives an example of a shot recorded closer to the net than it really was.
  • Different models give different answers. MoneyPuck, Natural Stat Trick, Evolving-Hockey and others all build their own models. Compare like with like, and don’t mix sources in one table.

How to use xG without fooling yourself

  1. Start with five-on-five. Special teams have their own shot patterns and much smaller samples.
  2. Use shares, not raw totals. A team with 50 xGF is great if it allowed 34 (Carolina) and ordinary if it allowed 50.
  3. Be patient with samples. In our view, team xG starts to mean something after roughly a quarter of the season, which is about where we are now. Player-level shooting gaps need much longer.
  4. Pair it with PDO. If a team’s xGF% is strong but its record is poor, check its shooting and save percentages (see our PDO guide). Utah is a current example: a 56.7 xGF% at five-on-five, but outscored 41-37.

Quick FAQ

Is xG better than Corsi? For judging chance quality, yes. For stability in small samples, Corsi still holds up well. Use both.

What is a good xGF%? Above 52 to 53 percent at five-on-five is a strong team, and above 55 percent is elite. So far in 2026-27, only Carolina, New Jersey, Utah and Tampa Bay are above 55 by MoneyPuck’s model.

Does the NHL publish xG? NHL.com publishes Corsi (SAT) and Fenwick (USAT) percentages. Public xG mostly comes from independent sites such as MoneyPuck and Natural Stat Trick.

Sources

  1. MoneyPuck: About and How It Works: model features, training sample, AUC, shooting-talent adjustment
  2. MoneyPuck: 2026-27 team (5v5) and skater (all situations) expected-goals data, pulled Nov. 12, 2026
  3. NHL stats: team SAT% / USAT% / 5v5 percentages
  4. NHL API: Evan Bouchard career shooting
  5. Wikipedia: Analytics in ice hockey: Corsi/Fenwick definitions and NHL names (SAT/USAT)