Nuclear explosion monitoring is commonly performed at regional or teleseismic distances, and consequently the spectra of the recorded seismic waves are significantly affected by the propagation path. Study of the explosion‐source signature requires separation of the source and the propagation effects. In this article, we analyze the data from two chemical explosions (54 kg and 123 kg of ammonium nitrate and fuel oil) conducted in central New Hampshire. We use small calibration explosions (0.454 kg Composition B boosters) to remove the propagation effects and to extract the source time functions (STFs) for the larger explosions using the empirical Green’s function method. Comparison of the deconvolved signals with the Mueller–Murphy model (Mueller and Murphy, 1971) shows that the observed STFs have significantly longer duration than the characteristic time scale predicted by the model. Source‐function period increase can be attributed to differences between nuclear and chemical sources. Alternatively, it can be explained by nonlinear attenuation effects near the source of shallow explosions.