Formal methods have been making inroads into the engineering of human–automation interaction (HAI) by allowing engineers to use mathematical proofs to determine whether normative or unanticipated erroneous human behavior can ever cause problems. However, these approaches are limited because they do not give engineers a way to assess the relative likelihood of different outcomes. In this work, we address this shortcoming by defining a new method that combines formal methods with human reliability analysis and probabilistic and statistical model checking. This approach ultimately allows analysts to compute the probability of different outcomes occurring in reactive HAI systems. We describe how this method was realized, assess its scalability, and demonstrate its capabilities with an automated teller machine example.