Discrete-choice models provide a simple way of representing utility-maximizing labor supply decisions in the presence of highly nonlinear and possibly non-convex budget constraints. Thus, it is not surprising that they are so extensively used for ex-ante evaluation of tax-benefit reforms. The question asked in this paper is whether it is possible and desirable to get still more flexibility by relaxing some of the usual constraints imposed on household preferences and rationality. We first suggest a model which attains flexibility by making parameters vary freely across hours choices. By embedding the traditional structural approach in this specification, it is shown that the restrictions on underlying well-behaved leisure-consumption preferences are rejected. More fundamentally still, the standard approach, i.e., the assumption of unitary households optimizing statically, is strongly rejected when tested against a general model with price- and income-dependent preferences. In a static environment, the result boils down to a rejection of the unitary model. Interestingly, restrictions from both structural and standard models also imply important discrepancies in estimated elasticities and simulated predictions of responses to a tax reform. In particular, large differences appear between standard models and the general model which possibly encompasses several interpretations including dynamic aspects and intrahousehold negotiation. These findings illustrate the difficulty to conduct policy analysis in a way which reconciles the best explanatory power and a framework consistent with economic theory. The general model we suggest may provide future research with an interesting setting to test some of the dimensions of household behavior.