For repeated events, fixed-effects regression methods—which control for all stable covariates—can be implemented by doing Cox regression with stratification on individuals. For nonrepeated events, we consider the use of conditional logistic regression to estimate fixed-effects models with discrete-time data. Known in the epidemiological literature as the case-crossover design, this method fails when any covariate is a monotonic function of time. Hence, no control for time itself can be included, leading to potentially spurious estimates. As an alternative, we consider the case-time-control method for estimating the effect of a dichotomous predictor. This method allows for the introduction of a control for time by reversing the role of the dependent and independent variables. In contrast to earlier work, we show that the method can be applied to data that contain only uncensored cases, and that it is possible to control for additional covariates, both categorical and quantitative. Simulation studies indicate that the case-time-control method is substantially superior to the case-crossover method and to conventional logistic regression. The methods are illustrated by estimating the effect of a wife's death on the hazard of death for her husband.