When Do NHL Teams Pull the Goalie? Empty-Net Goals in 3,936 Games

Almost everyone does it. Of the 1,308 times a team trailed by one goal with three minutes left in regulation, 94.4% pulled the goalie, out by a median of 1:49 left. Down two at the same point (802 times), 92.0% pulled and the median came 56 seconds earlier, at 2:45.

NHL play-by-play, regular season. “Pull time” is the first recorded play with the net empty: the goalie was out by then, and the actual pull came a median of about 20 seconds earlier (the gap to the previous play), so every pull time here is the latest it could have been. Extra-attacker changes on a delayed penalty are not counted. Down-1 and down-2 spots are teams trailing by that much with exactly 3:00 left in the third, and their first pull in the last four minutes.

The one trend is down two: the median pull moved from 2:39 to 2:54 left in two seasons, and the share pulled with more than three minutes to go rose from 32.6% to 45.2%. Down one, the timing has barely moved.

When a team trailed by one goal with three minutes left, this is how the game ended:

The three rows add to 100%. “Tied it” counts a team that tied the game at any point after the 3:00 mark, including ties later lost in overtime, a shootout or to another goal, unless it still lost by two or more.

So the trailing team got the game tied about one time in five, and gave up an empty-net goal in 42.0% of these games, about four in ten. Down two, the numbers are bleaker: 1.9% tied it and 88.5% lost by two or more.

That still makes sense for the team. A one-goal loss and a two-goal loss earn the same zero points in the standings, so a coach gives up nothing in the standings by risking the empty-netter. The bettor holding the trailing team at +1.5 does care: for them the tying goal adds nothing and the empty-netter loses the bet.

The 73 teams that never pulled tied it 71.2% of the time — not because staying in net is a better plan, but because many of them tied the game before they needed to pull.

Trailing by one with three minutes left, split by when the goalie came out:

Earlier pulls come with more goals at both ends: the extra attacker scores more often, and so does the other team. The rate of getting the game tied moves much less than either. Be careful reading cause into it, though — a coach pulls early when the moment looks right (an offensive-zone faceoff, sustained pressure), so the early group is not a random sample.

Across all 3,936 regular-season games, 34.4% had at least one empty-net goal — 1,479 in all, 0.38 a game and 6.1% of every goal scored. Nearly all of them (96.1%) came in the last four minutes of the third period.

Final margins split three ways: 39.9% of games were decided by one goal (overtime and shootout games, 22.1% of all games, are all one-goal games), 19.9% by exactly two, and 40.3% by three or more. The two-goal games are where empty-netters live: in 64.3% of them an empty-net goal had turned a one-goal lead into two (71.7% in 2025-26).

That is exactly the line between the favorite -1.5 winning and the underdog +1.5 cashing. From the trailing team's side, a one-goal deficit with three minutes left still covers +1.5 about 58% of the time (57.7%: everything except the 42.3% that lost by two or more), and 42.0% gave up an empty-net goal.

Totals feel it too. An empty-net goal was what pushed a game over 5.5 in 6.7% of all games, and over 6.5 in 5.2%.

None of this is hidden from pregame numbers. A puck line or total priced on final scores — the book's or ours — already has these late goals inside it. Our NHL puck line rule (the road team +1.5 when our model gives it 65% or more) is graded on final scores, empty-netters included, on the NHL predictions page.

In any one season, it looks like it: in 2025-26 the median pull time, down one, of the 27 teams with at least 10 pulls ranged from 2:30 to 1:24 left. But a team only finds itself trailing by one with three minutes left about 14 times a season, and the habit barely carries over. Comparing each team's median pull time with the next season's, the correlation was 0.26 and then 0.29.

We also tested it the way a bettor would use it: rank teams by last season's pull timing, then look at this season's late one-goal games. Last season's early pullers lost by two or more 45.6% of the time, the middle group 40.2% and the late pullers 44.5% — no order at all. A coach's reputation for pulling early is not something to bet on.

The useful part is the base rate. When a team trails by one with about three minutes left, its +1.5 has held about 58% of the time and the leader's -1.5 about 42%; down two, the trailing team has tied it only 1.9% of the time. Set that next to the live price you are being offered and you know whether the price assumes more or less than history does.

It is a base rate for the average game, not a forecast for this one: the score, who has the puck, a power play or a goalie already pulled all move the next two minutes, and the live line moves with them. For how the whole puck line market works, see NHL betting analytics.