Discrete-choice models of labor supply have become very popular for ex ante evaluations of policy reforms as they easily account for non-convex budget sets. We test the constraints imposed in practice on these models and suggest a fully flexible model that significantly improves fit