Satellite estimates of burned area, associated carbon monoxide (CO) emission estimates, and CO column retrievals do not agree on the peak fire month in Africa, evident in both Northern and Southern Africa though distinct in the burning seasonality. Here we analyze this long‐standing problem using 1) a top‐down Bayesian inversion of MOPITT CO columns during 2005-2016 into surface CO emissions, and 2) the bottom‐up GFED 4.1s fire emission database. We show that GFED 4.1s underestimates CO emissions by 12-62% in the late fire season and hypothesize that this is partly because it assumes seasonally‐static emission factors. However, the degree to which emission factors would have to vary through the season to bring top‐down and bottom‐up in agreement cannot be confirmed by past field‐based measurements. Improved observational constraint on the seasonality of burned area, fuel combustion, and emission factors would further reduce the discrepancy between bottom‐up and top‐down emission estimates.