The literature has come to no agreement about the empirical validity of the so-called weak government hypothesis. According to this hypothesis, political fragmentation should lead to higher government expenditure. With the aim of reconciling the empirical evidence with theory, in this paper we discuss and test a new hypothesis about this relationship: that fragmentation should matter for public spending only to the extent that the degree of polarization is high enough. Our results for a sample of presidential democracies show that a marginal change in the level of fragmentation in the governing coalition affects positively the size of the budget, but only if there is some degree of polarization. We also find that what matters for fiscal policy in presidential democracies is the degree of fragmentation and polarization within the governing coalition, rather than in the legislature at large. For parliamentary democracies we find erratic patterns for the relationship between fragmentation and public spending. Our results suggest interesting differences between presidential and parliamentary systems.