This paper uses state-level data to estimate the effect of federal defense spending shocks on state real activity. We find moderately strong evidence that for the average state the fiscal multiplier is larger during recessions. However, there is substantial heterogeneity across the cross-section. The degree of non-linearity in the effect of spending shocks is larger in states that are subject to a higher degree of financial frictions and lower labour market rigidity. In contrast states with a prevalence of mining and agricultural industries tend to have multipliers that are more similar across business cycle phases.