We apply an inverse method to estimate the carbon monoxide emissions in Atlanta, Georgia. The resultant carbon monoxide inventory is unrealistically characterized by temporal oscillations on the frequency of 1 hour. We suspect that the difficulty in deducing the emissions is due to inhomogeneities in the spatial distribution of the emissions. In several controlled experiments we reproduce the oscillations by introducing relatively small errors into the spatial distribution of the emission inventory. In similar experiments with isoprene in which the emissions are more homogeneous, we do not find this problem. These results are discussed in the context of previous inverse studies of carbon monoxide and isoprene.¿ 1997 American Geophysical Union