This letter evaluates the performance of auxiliary regression-based specification tests for parametric duration models estimated with censored data. The test using asymptotic critical values has poor size. Bootstrapping corrects the size problem but results in a biased power curve.