Empirical models of labour supply adopting the collective approach have commonly used the decentralized representation and a reduced form specification of the sharing rule. This procedure has two crucial drawbacks that in principle make it inappropriate for the very same type of applications that are thought to be mostly relevant for this family of models, i.e. tax reform simulations. The first problem concerns the decentralized representation. The possibility of decentralizing the maximization of household welfare rests on the convexity of the budget sets. However, both the actual tax systems and the tax reforms might imply significant non-convexities: this makes the decentralized representation in general inappropriate both for estimation and for simulation. The second problem concerns the specification of the sharing rule, which typically is not a structural one, but rather a reduced form, e.g. a combination of exogenous variables (wage rates, unearned incomes etc.). Such a specification might provide a reasonable approximation to the current intra-household allocation choices under the current tax rule, but in general it cannot be used for simulating the effects of a different tax rule. Analogously, the sharing rule in general will be different depending on whether both partners work or not. We propose - and illustrate with some preliminary results - a model that permits the estimation of a structural sharing rule.