Theories of democratic governance posit that citizens should reward politicians for good service and punish them for bad. But does electoral accountability work as theorized, especially in developing country contexts? We contribute a new empirical answer to this question. We study Southern African democracies, where infrastructural investment in basic services has expanded widely, but not universally. We first analyze whether increases in local service provision predict increases in voting returns at various levels of geographic aggregation. We then analyze geo-coded, individual-level survey data that allow us to consider the relationship between service delivery, political attitudes and voting intentions. Together, these analyses demonstrate, quite unequivocally, that citizens do not reward good service. We explore the mechanisms that may account for this surprising finding.