This paper provides an analysis of potential unequal burden of bribery in schools on poor households in developing countries. The rich are more likely to pay bribes in the standard model where the probability of punishment for bribe taking by a teacher is the same irrespective of income of the household. This model is, however, not appropriate in the context of a developing country lacking in rule of law, where the ability to punish a corrupt teacher depends on a household’s economic status. Bribery is likely to be regressive at the extensive margin in this case. The conditions required for progressivity at the intensive margin are also quite stringent. A signifcant part of the available empirical evidence, however, finds bribes in developing countries to be progressive, thus contradicting the theoretical predictions above. We argue that this conflict may largely be due to the identification challenges arising from ability and preference heterogeneity. Using ten year average rainfall variations as instrument for household income in rural Bangladesh, we find that corruption is doubly regressive: (i) the poor are more likely to pay bribes (income elasticity [-0.73, -1]), and (ii) among the bribe payers, the poor pay a higher share of their income. The IV results for intensive margin are in contrast to the OLS estimate that shows bribes to be increasing with household income, substantiating the worry about spurious progressive effects. The results imply that ‘free schooling’ is free only for the rich, and corruption makes the playing field skewed against the poor.