Established authorities have disagreed about the relative importance of the upper and lower atmosphere in producing astronomical refraction for nearly two centuries. This paper resolves the problem and corrects some prominent errors. The refraction near the horizon is explored in some detail, and its relation to terrestrial refraction, and the effects of thermal inversions in the nocturnal boundary layer, are examined. At many observatories, the refraction at the apparent horizon comes mostly from the air between the observatory and sea level.