In airline revenue management, an accurate prediction of cancellations is crucial, since a significant number of bookings are cancelled before departure. Accurate estimation of cancellation behavior is essential for airlines, so that they can allow more reservations on a flight than there is physical capacity (“overbooking”), which is a significant source of revenue. Cancellation probabilities depend in a complex manner on several flight-related and passenger-related attributes. A proportional hazard model is applied to predict the “hazard rates”, i.e. the conditional risks of a reservation being cancelled. These are used to forecast the expected number of cancellations depending on bookings on hand and forecasted future demand. We enhanced the standard maximum-likelihood estimator in order to obtain practicable processing time and memory consumption. The new method provides stable predictions and improves accuracy significantly compared to a time series approach.