Authors of benefit-incidence analyses (BIA) have to impute subsidies using assumptions about the relationship between unobserved subsidies ‘captured’ by the household and what can be observed at the household and aggregate levels. This paper shows that one of the two assumptions used in BIA studies to date will necessarily produce a more pro-rich (or less pro-poor) picture of government health spending than the other, depending on whether utilization is more pro-rich or pro-poor than fees paid to public providers. Both assumptions have their disadvantages, and the paper suggests a couple of alternatives that explicitly link fees paid to the costliness of care. It shows that in the most likely case where fees are distributed in a more pro-rich fashion than utilization, the two traditional assumptions will produce less pro-rich distributions of subsidies than the two new alternatives. Also considered are three complications that arise in BIA studies, including factoring in social health insurance. The paper's theoretical results are illustrated with an empirical BIA for Vietnam. Copyright © 2011 John Wiley & Sons, Ltd.