How Much of a Goalie’s Save Percentage Is the Goalie?

The question: skill, opportunity or the team?

Save percentage is the stat everyone uses to grade a goalie. But a goalie only faces the shots the team allows. A team that keeps opponents to the outside hands its goalie easy saves. A team that gives up odd-man rushes and cross-ice passes does the opposite.

So how much of a goalie’s save % is really the goalie? The cleanest test is to watch the same goalie play for two different teams. If save % is skill, it should travel with the goalie. If it is the team, it should stay behind.

We used every NHL regular season from 1955-56 to 2025-26: 127,409 goalie games from the NHL’s own game logs. Since 2017-18 we can also split each shot into its expected goal value, using our play-by-play expected-goals (xG) model. Throughout, “1 point” of save % means .001, so .905 to .910 is 5 points.

What happens when a goalie changes teams

We found 1,551 cases of a goalie playing 500+ shots for the same team in back-to-back seasons, and 321 cases of a goalie doing it for two different teams. For each, we compared save % above the league average in year one and year two.

Across all seasons, a goalie who stayed kept about 0.40 of the gap from league average into the next year (90% range 0.33 to 0.45). A goalie who moved kept 0.07 (−0.03 to 0.16). We use a slope here, not a raw correlation, because teams pick which goalies to acquire based on their save %. That kind of selection does not bias a slope.

One era breaks the pattern: from 2005 to 2014, movers carried over as much as stayers. That is 63 moves, and the range is wide. Every other era shows the same gap.

Famous moves fit the pattern

• Denis Herron, Montreal to Pittsburgh (1981-82 to 1982-83): 38 points above league average to 38 points below, the biggest swing in our data.
• Philipp Grubauer, Colorado to Seattle (2020-21 to 2021-22): 15 points above average to 18 below. The shots faced were no harder by xG; the save % simply fell.
• Darcy Kuemper, Washington to Los Angeles (2023-24 to 2024-25): 13 below average to 21 above.
• Filip Gustavsson, Ottawa to Minnesota (2021-22 to 2022-23): 15 below to 27 above.
• Dominik Hasek, Buffalo to Detroit (2000-01 to 2001-02): 17 above to 8 above. Still great, but half the gap was gone.

You can tell a story about each one. The point of the full sample is that, on average, the stories cancel. Knowing a goalie’s save % on the old team tells you almost nothing about save % on the new one.

Goalie, team or luck: splitting the spread

Movers are the cleanest test, but they are rare. To use every season, we fitted a model with four parts: the goalie’s career level, the goalie’s year-to-year form, the team’s level and the team’s year-to-year form. Partners on the same team (the starter and backup face the same defence) and goalies who changed teams let the model tell goalie and team apart.

Luck is not guessed. With 1,500 shots and a .905 average, pure chance moves a save % by about 7.5 points either way (one standard deviation). That is a coin-flip fact, so we removed it before fitting the rest.

In points of save %, a goalie one standard deviation better than average is about 4 points better over a career, plus or minus 3 points of form in any one year. The team effect is about as big: 3 points of lasting team level plus 3 points of year-to-year swing.

The luck share has grown because goalies have converged. The gap between good and bad goalies has narrowed since the 1970s, while the luck from 1,500 shots stays the same. So the same season of results now tells you less about the goalie.

Shot quality: the team’s fingerprint

Since 2017-18 we can split save % into two pieces. Expected save % is what a league-average goalie would stop given the shots faced (location, type, rebounds). Goals saved above expected (GSAx) is what the goalie did beyond that. We work per unblocked shot, because xG covers missed shots too.

Expected save % is a team stat, plainly. Our model puts 47% of its spread on the team and only a little on the goalie. Goalies have a small say in it: rebound control and puck handling change the next shot.

The surprise is GSAx. It is meant to be the goalie-only number, yet for movers it did not carry over either, and it lined up with the new teammates (0.23). Our model still assigns about 20% of GSAx spread to the team. Location-based xG misses things the team controls: screens, pre-shot passes and rush chances. Public xG models share that blind spot.

So the theory that teams shape save % holds, with a twist. Teams do hand their goalies easier or harder shots, and the part xG can see is only some of it. The rest hides inside the “goalie” numbers.

Can you forecast a goalie on a new team?

Here is the practical test. For every season from 1970-71 on, we fitted our models only on the 10 seasons before it, then forecast every goalie with 500+ shots. We scored each forecast by how much of the predictable spread it caught. Pure luck cannot be predicted by anyone, so we took that out.

For a goalie on a new team, the best save % forecast is the league average, nudged toward the new team’s shot quality. Last season’s raw save % is worse than useless; it is an anchor that pulls the forecast the wrong way. We break down good-to-bad and bad-to-good moves in what happens when a good goalie joins a bad team, and rank every franchise in the best and worst NHL teams for goalies.

What it means for fantasy hockey

Fantasy leagues score goalies on wins, goals-against average (GAA), save % and shutouts. We checked how each repeats from one season to the next for goalies who stayed and goalies who moved (1983-84 on, 800+ shots, one team all season).

• Draft the team, then the goalie. GAA and wins repeat more than save % for goalies who stay, because the team stays too. When a goalie moves, project those categories from the new team.
• Fade career years on new teams. A goalie who signs somewhere new after a big save % season should be projected close to league average, adjusted for the new team’s shot quality.
• Buy the shot-suppressing team’s goalie, not the name. Expected save % repeats for teams at 0.55 year to year, much better than any goalie number. That is the most stable piece of the puzzle.
• Workload is mostly fine. Starters playing the second night of a back-to-back showed no clear save % drop (450 starts since 2017-18; the range runs from 1 point worse to 4.5 points better). Coaches pick those spots, so this is not a free pass.

Check tonight’s starting goalies and the season numbers on our goalie stats page, which show expected save % and GSAx side by side.

Does the betting market get goalies wrong?

If save % is partly the team, a market that leans on save % could overrate goalies on good defensive teams. We tested it on near-close moneylines for 2,618 games in 2024-25 and 2025-26, using only goalie numbers known before each game.

• Goalie quality is priced. Adding the starters’ GSAx, raw save %, or the gap between them (the “team inflation”) to the closing price made 2025-26 predictions slightly worse, not better. No term was close to significant.
• Backups are priced too. Teams starting their backup won 48.1% against 48.9% priced (1,935 team-games; the gap’s 90% range is −2.5 to +1.1 points). Betting every backup returned −4.9% at median prices.
• The starter-backup gap is small. By our shrunk GSAx, the average starter is 0.08 goals a game better than the backup. That is worth about 1.3 points of win probability, close to what the market already charges.

One thread is worth watching. On the second night of a back-to-back, a fresh goalie (usually the backup) stopped more than usual: 2.4 points of unblocked save % above expected, and it held in both halves of the data. Yet those teams won 3.2 points less often than their closing price (764 team-games). That matches our back-to-back study. The tired skaters are the problem, not the backup goalie.

Betting against those tired teams returned +2.6% at median prices, but the 90% range runs from −3.0% to +8.5%, and the second season was −0.9%. That is a watch-list note, not a bet.

What we could not settle

• Movers are not random. Teams trade for goalies after good years and dump them after bad ones. Slopes handle selection on last season, and we matched movers to similar goalies who stayed (same save %, workload, age and era), but aging and injuries still sit in the mix.
• Before 2017, team shot quality is invisible. We only have shot counts. Play-by-play coordinates exist from 2009-10, and pulling them would extend the xG test by eight seasons.
• Rink scorers do not explain it. A generous home scorer could pad saves for both goalies in a building. We checked: the shared building effect on save % is about zero in every era. The team effect shows up on the road too.
• Two seasons of odds are not enough to rule out a mispricing smaller than about 2 points of win probability.

Show the work

Frequently asked questions

More in Speed, faceoffs, lines and goalies

• Verdict: Promising

  Do Faster NHL Teams Win More? What NHL EDGE Speed Data Says

• Verdict: Myth

  Does Playoff Hockey Reward Hitters Over Speed? We Tested Five Postseasons

• Verdict: Mixed

  NHL Faceoff Matchups: Ignore the Head-to-Head, Watch the Hands

• Verdict: Mixed

  NHL DFS Stacking Works. “Chemistry” Is Mostly Noise.

• Verdict: Mixed

  The Best and Worst NHL Teams for Goalies: Boston Helps Most, Philadelphia Hurts Most

• Verdict: Mixed

  When a Good Goalie Joins a Bad Team for Goalies, Save % Drops About .005. The Team Explains Part of It, and Luck Plays a Part

All Prop Lab research →

Keep reading

• Best and worst NHL teams for goalies →Every franchise rated by how the same goalies did there and elsewhere, from 1967 to today.
• Good goalie, bad team for goalies →What happens to save % when a goalie moves between good and bad teams for goalies, and whether it is bettable.
• Betting NHL teams on back-to-backs →How much tired teams lose, the backup-goalie split, and whether the price already knows.
• NHL Prop Lab: hockey betting edges and myths →The full hockey research hub: what survived testing, what failed, and what we are tracking live.

Sources and method

1. NHL stats (NHL.com)
2. NHL API reference (community documentation)
3. The Odds API (historical NHL moneylines, puck lines and totals)
4. Empirical Bayes shrinkage
5. Responsible Gambling Council (Canada)
6. ConnexOntario (1-866-531-2600)
7. 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.