Recognizing that current theories of team learning do not apply to short-term action teams, we conceptualize how action teams may learn, and test hypotheses regarding the performance-related effects of such learning, the mechanisms mediating such effects, and the conditions governing their magnitude. We operationalized the level of Action Team Learning (ATL) in terms of the regularity and number of role-based, guided team reflexivity experiences of an action team's members. Testing our hypotheses on 250 surgical teams, we find that higher levels of ATL are associated with shorter surgical duration, with this effect mediated by team helping and workload sharing, particularly under conditions of greater team task complexity. Additionally, we find higher levels of ATL to be directly associated with a reduced number of adverse events in low-complexity surgeries.