This paper investigates how well regulator examinations predict bank failures, and how best to incorporate examination information into an econometric model of time-to-failure. We estimate proportional hazard models with time-varying covariates and find that examiner ratings help explain the failure hazard. Both the overall rating of a bank's condition and management, i.e., the composite CAMELS rating, and ratings of specific components contain information. In addition, we find that the marginal "effect" of ratings is non-linear, in that the impact of a rating downgrade on the probability of failure is larger, the weaker a bank's initial rating.