Impulse responses to government spending shocks in Standard Vector Autoregressions (SVARs) typically display "expansionary" features. However, SVARs can be subject to a "non-fundamentalness" problem. "Expectations - Augmented" VARs (EVARs), which use direct measures of forecasts of defense spending, typically display "contractionary" responses to a defense news shock. I show that, when properly specified, SVARs and EVARs give virtually identical results. The reason for the widespread, opposite view is that defense shocks have "contractionary" effects while civilian government spending shocks have "expansionary" effects. Existing EVARs and SVARs, however, include only total government spending. In addition, the former are typically estimated on samples that include WWII and the Korean war, when defense shocks prevailed, while the latter are estimated mostly on post-1953 samples, when civilian shocks prevailed.Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.