This paper studies the design of couples' income taxation when consumption and labor supply decisions within the couple are made by maximizing a weighted sum of the spouses' utilities; bargaining weights are given but specific to each couple. Information structure and labor supply decisions follow the Mirrleesian tradition. However, while the household's total consumption is publicly observable, the consumption levels of the individual spouses are not observable. With a utilitarian social welfare function we show that the expression for a spouses' marginal income tax rate includes a "Pigouvian" (paternalistic) and an incentive term. The Pigouvian term favors a marginal subsidy (tax) for the high-weight (low-weight) spouse, whose labor supply otherwise tends to be too low (high). The sign and the magnitude of the incentive term depend on the weight structure across couples. In some cases both terms have the same sign and imply a positive marginal tax for the low-weight spouse (who may be female) and a negative one for the high-weight spouse (possibly the male). This is at odds with the traditional Boskin and Sheshinski results. Our conclusions can easily be generalized to more egalitarian welfare functions. Finally, we present numerical simulations based on a calibrated specification of our model. The calculations confirm that the male spouse may well have the lower (and possibly even negative) marginal tax rate.