The receiver-to-source backazimuth of atmospheric infrasound signals is biased when cross-winds are present along the propagation path. Infrasound from 598 surface explosions from over 30 years in northern Finland is measured with high spatial resolution on an array 178 km almost due North. The array is situated in the classical shadow-zone distance from the explosions. However, strong infrasound is almost always observed, which is most plausibly due to partial reflections from stratospheric altitudes. The most probable propagation paths are subject to both tropospheric and stratospheric cross-winds, and the wave-propagation modelling in this study yields good correspondence between the observed backazimuth deviation and cross-winds from the European Centre for Medium-Range Weather Forecasts Reanalysis (ERA)-Interim reanalysis product. This study demonstrates that atmospheric cross-winds can be estimated directly from infrasound data using propagation time and backazimuth deviation observations. This study finds these cross-wind estimates to be in good agreement with the ERA-Interim reanalysis.