It is generally accepted that government health expenditures should disproportionately benefit the poor. And yet in most developing countries the opposite is the case. This paper examines the implications of a central assumption of benefit incidence analysis, namely that the unit cost of a government-provided service bears no relation to the out-of-pocket payments paid by the patient. It argues that a more plausible assumption is that larger out-of-pocket payments for a given unit of utilization reflect more (or more costly) services being delivered. The paper compares - theoretically and empirically - the standard constant-cost assumption with two alternatives, namely that the cost of care in a specific episode of utilization is (a) proportional to or (b) linearly related to the amount of money paid out-of-pocket by the patient. An interesting special case of the linear relationship is where subsidies are focused on a basic unit of care and additional costs are met dollar-for-dollar by additional fees. The paper shows that if fees are more pro-rich than utilization, government spending will be least pro-rich under the constant-cost assumption and most pro-rich under the proportionality assumption. The linear assumption results in a concentration index for subsidies that lies between these two extremes. These results are borne out in an analysis of the incidence of government health spending in Vietnam (a country where fees are more pro-rich than utilization); indeed, under the constant-cost assumption, subsidies are pro-poor while they are pro-rich under the proportionality assumption. The paper also considers the biases created by not allowing for insurance reimbursements.