The nested or sequential logit model is the only computationally tractable random-utility model that permits correlation among the random components of the utility functions of different alternatives. In this paper, two specification tests are described for nested logit models. One is a test of a maintained model against a nonnested alternative. This test can be used, among other purposes, to discriminate among models with different tree structures. It can be implemented by hand using the results of sequential estimation of the models under consideration. The other test consists of comparing the sequentially estimated parameter values with values produced by an asymptotically efficient estimation technique. This test does not require estimating an alternative to the maintained model.