Résumé : Having accurate poverty statistics is of primary importance for both researchers and policy-makers. Based on original data on Burundi, we investigate the sensitivity of poverty headcount calculations to methodological choices along two dimensions: (1) the vector of prices that is used to value consumption, and (2) the level of consumption that is considered (household versus individual). We find different numbers of poor when these dimensions are accounted for. In our context, it translates into a larger number of poor when country-level instead of local prices are used, and when intra-household allocation of consumption is accounted for – the latter being mainly driven by poor children living in households considered as non-poor.