Save percentage is the most-quoted, least-useful goalie stat
Save percentage is the official, league-sanctioned, broadcast-graphic- approved way to evaluate an NHL goaltender. It appears on every Hockey Night graphic, every back-of-card stat block, and almost every contract negotiation. It is also wrong about goalies in some specific and consistent ways, and it has been wrong about them for as long as anyone has used it. The problem is not that the math is bad. The problem is what the math is hiding.
Save percentage answers exactly one question: of the shots that hit the net, what fraction did the goalie stop? A .920 goalie stopped 92 out of every 100. The number is clean, intuitive, easy to compare across seasons, and easy to slap on a graphic. It is also, on its own, useless for the thing most people use it for, which is figuring out who's playing well in net.
The shots aren't the same
The fundamental flaw is in the denominator. Save percentage treats a shot from the slot as identical to a shot from the blue line. It treats a one-timer on a two-on-one identically to a wrap-around from a bad angle that hit the post and dribbled in. A goalie who plays behind a defensive structure that funnels every attempt to the outside is going to post a great save percentage by stopping a long sequence of shots that mostly weren't going in anyway. A goalie who plays behind a team that bleeds chances from the high-danger area is going to look mediocre while making a series of saves that statistically should have been goals.
In other words: save percentage is a measure of two things — the goaltender, and the defense in front of them — collapsed into one number, with no way to separate them. The graphic might as well say "save percentage and team-defense quality, mixed together, rounded to three decimals."
The leaderboard problem
Look at any season's save-percentage leaderboard and you'll find a pattern: a meaningful chunk of the top names are goalies on well-structured defensive teams. Some of those goalies are excellent. Some are decent goalies playing in a system that flatters them. The leaderboard does not distinguish.
The same effect runs in the other direction. There are reliably good goalies whose save percentages look pedestrian for years because they spent those years on rebuilding teams that gave up forty quality chances a night. Hockey people who watched the games knew. Whoever was reading the back of the card did not.
Sample-size noise
Even setting aside the shot-quality problem, save percentage is unstable across small samples in a way that the precision of the number obscures. A goalie playing twenty games is in noise territory. A goalie playing forty games is in less noise but still meaningful noise. Even an entire 60-start season has enough variance in shot quality faced that a "true" .915 goalie can post anything from .908 to .922 without being a different goalie at all.
We rarely think about save percentage that way because it's reported to three decimals and feels precise. The third decimal is almost entirely noise. The second decimal is partially noise. The broadcast graphic that tells you Goalie A is at .918 and Goalie B is at .913 is reporting a difference that, in most samples, isn't statistically meaningful.
What's better
The fix is not to throw save percentage out — it's still the cleanest rough cut on the goalie position — but to read it alongside metrics that handle the shot-quality problem. The two that have stuck:
Goals saved above expected (GSAx). The shot-quality adjustment. A model assigns each shot a probability of becoming a goal based on its location, type, and game context. A goalie's GSAx is the number of goals they saved relative to what an average goalie would have allowed on the exact same set of shots. Above zero is good. Above ten over a full season is excellent. Above twenty is a Vezina case. The metric is not perfect — the shot-quality model is only as good as its inputs — but it answers the question save percentage was supposed to answer.
High-danger save percentage. A simpler version of the same idea: save percentage restricted to shots from the most dangerous areas of the ice. League average sits well below regular save percentage because the shots are harder. Goalies who are good in close are fundamentally different evaluators than goalies who pad numbers on perimeter pucks.
Both stats are routinely available now, on Natural Stat Trick and Evolving-Hockey and a handful of other public sources. They aren't broadcast-friendly, which is most of why TV still leans on save percentage. They are dramatically more informative if you want to know who's actually playing well.
The honest version of the rule
Save percentage is an okay stat for fans and a bad stat for evaluators. It will reliably tell you the difference between a replacement-level goalie and an elite one over a long career. It will not reliably tell you who's playing better this month, who deserves the second-half starts, who to ride into a playoff series, or which goalie's contract is worth what. For any of those questions, the broadcast number is the wrong tool.
None of this is news inside hockey operations. Every front office of the last decade has built or bought a model that adjusts for shot quality. The reason it still matters to point out is that the public conversation — fan debates, podcast takes, contract analyses written for a general audience — still leans on save percentage as if it were the answer. It is the headline. The analysis is somewhere underneath.
The next time a graphic flashes up showing two goalies separated by eight thousandths of a save percentage, remember what those eight thousandths represent: a difference smaller than the noise in the sample, calculated from a denominator that pretends every shot is the same. The number is not a verdict. It's a starting point.