We use dense seismic network recordings of accurately located surface explosions in northern Utah to shed light on the propagation of infrasound through the stratosphere. The data clearly show propagation of infrasound downwind from the source, as expected. The data also clearly show the penetration of infrasound into geometric shadow zones near the source and the spread of infrasound to a distance of 800 km from the source. The spread of infrasound both toward and away from the source is not predicted by applying either ray theory or the full‐wave finite‐difference technique to smooth ground‐to‐space (G2S) models. The mismatch between synthetics and data suggest a missing component in these models, possibly a small‐scale gravity‐wave structure. Comparison of the network recordings of approximately 1500 infrasound signals with travel‐time predictions based on rays shows no significant average bias in the travel times. On average, recorded signals arrived 1 s earlier than predictions. Travel‐time residuals are normally distributed about the mean with a standard deviation of 15 s. The small bias of the travel‐time predictions indicates that despite the fact that small‐scale structure is averaged out of commonly used G2S models, the large‐scale structure of the atmosphere is accurately represented. The scatter of travel‐time residuals is suggestive of small‐scale structure missing from the models that we used to make the predictions, but firm conclusions would require a more in‐depth study.