Despite significant improvement in female schooling over the last two decades, only a small proportion of women in South Asia are in wage employment. We revisit this puzzle using a nationally representative data set from Bangladesh. Probit regression results show that even after accounting for human capital endowments, women are systematically less likely to participate in paid work than men. Oaxaca decomposition of the gender gap confirms that most of it (i.e. 95%) is unexplained by endowment differences. Instead, community norms such as the practice of purdah (i.e. female seclusion) have a statistically significant and negative effect on women's participation in paid work. We do not find any evidence that purdah norm variable affect paid work participation indirectly, via determining the labor force participation decision. The correlation between current work participation and purdah norm in natal household is insignificant confirming that the result is not driven by omitted individual-specific socio-economic factors. We also use data on past purdah practice of the current community to estimate an instrumental variable Probit regression model and rule out the possibility of reverse causality. Detailed decomposition analysis reveals that community purdah norm accounts for a quarter of the total unexplained gap. The findings are robust to controls for the influence of co-resident in-laws, household structure, marital status, and a wide range of community characteristics such as ecological factors, presence of NGOs, provision of public infrastructure, remoteness and local labor market conditions including the norm of unacceptability of unmarried women's outside work in the community.