We take to the data an RBC model with two salient features. First, we allow government consumption to directly affect the marginal utility of consumption. Second, we allow public capital to affect the productivity of private factors. On the one hand, private and government consumption are estimated to be substitute goods. As a consequence, the estimated response of private consumption to a government consumption shock is negative, as in models with separable government consumption, but such response is much stronger. Further, substitutability makes labor supply to react less, so the estimated output multiplier is lower than in models with separabilities, peaking - on impact - at 0.39. On the other hand, non-defense public investment enhances mildly or negligibly, depending on the specification, the productivity of private factors. In those specifications where non-defense public investment is found to be productive, a non-defense investment shock generates the following estimated responses (after several quarters): a positive reaction for private consumption, Tobin’s q, private investment and real wages. Unlike models with unproductive government investment, the estimated output multiplier builds up over time, starting well below one on impact, then reaching 0.93 after three years and 1.44 after six.