Estimation of global atmospheric emissions from biomass burning has used a variety of procedures, most of which involve chain multiplication. The terms in the chain of inference are often poorly quantified, and in many cases there is reason to suspect bias and intercorrelation between terms. The leading estimates appear to have been conservative; however, uncertainty explodes when imprecisely known (coefficient of variation >0.3) terms are multiplied. Thus little confidence can be had in the precision of current best estimates. The situation can be improved through a combination of spatial disaggregation and process-oriented modeling.