Background: Despite a high reported incidence rate of concussion, little is known about the on-field characteristics of injurious head impacts in National Football League (NFL) games. Purpose: To characterize on-field features (location, player position, and time during game) and biomechanical features (anticipation status, closing distance, impact location and type) associated with concussions in NFL games over a 4-season period (2010-2011 to 2013-2014).Study Design: Descriptive epidemiology study. Methods: We analyzed video of a subset of reported, in-game concussions for the 2010-2011 to 2013-2014 seasons. These videos represented a sample of injuries that were diagnosed and reported on the same day and that could be clearly associated with an in-game collision as captured on video. We determined anticipation status, closing distance, impact location on the injured player’s helmet, and impact type (helmet-to-helmet, helmet-to-body, or helmet-to-ground). Associations between these variables were analyzed by use of descriptive statistics and tests of association. Results: A total of 871 diagnosed concussions were reported as occurring during NFL preseason, regular season, and postseason games for the 2010-2011, 2011-2012, 2012-2013, and 2013-2014 seasons. A total of 1324 games were played during this period, giving a concussion incidence rate of 0.658 per game (95% CI, 0.61-0.70). From the video-reviewed subset (n = 429; 49.3%), the majority of injurious impacts occurred with good anticipation (57.3%) and <10 yards of closing distance (59.0%). An association was found between anticipation status and play type (χ23 = 27.398, P < .001), largely because injuries occurring on pass plays were more likely to be poorly anticipated than injuries during run plays (43.0% vs 21.4%; χ21 = 14.78, P < .001). Kick returns had the greatest proportion of well-anticipated impacts (78%) and the greatest proportion of impacts with ≥10 yards of closing distance (80%). The type of impact was approximately equally divided between helmet-to-helmet, helmet-to-body, and helmet-to-ground types. The impact location was broadly distributed over the helmet of the injured player. Conclusion: In-game concussions in the NFL occurred through a diverse variety of mechanisms, surprisingly tended to be well-anticipated, and, also surprisingly, occurred with <10 yards of closing distance. The impacts causing concussion were broadly distributed over the helmet. More concussions occurred during the second half of game play, but we do not have evidence to explain this finding.