In the standard model of corruption, the rich are more likely to pay bribes for their children's education, reflecting higher ability to pay. This prediction is, however, driven by the assumption that the probability of punishment for bribe-taking is invariant across households. In many developing countries lacking in rule of law, this assumption is untenable, because the enforcement of law is not impersonal or unbiased and the poor have little bargaining power. In a more realistic model where the probability of punishment depends on the household's economic status, bribes are likely to be regressive, both at the extensive and intensive margins. Using rainfall variations as an instrument for household income in rural Bangladesh, this paper finds strong evidence that corruption in schools is doubly regressive: (i) the poor are more likely to pay bribes, and (ii) among the bribe payers, the poor pay a higher share of their income. The results indicate that progressivity in bribes reported in the earlier literature may be due to identification challenges. The Ordinary Least Squares regressions show that bribes increase with household income, but the Instrumental Variables estimates suggest that the Ordinary Least Squares results are spurious, driven by selection on ability and preference. The evidence reported in this paper implies that "free schooling" is free only for the rich and corruption makes the playing field skewed against the poor. This may provide a partial explanation for the observed educational immobility in developing countries.