The purpose of the present study is to examine whether microfinance reduces poverty in Bangladesh drawing upon the nationally representative household panel data covering 4 rounds from 1997 to 2005. A special attention was drawn to the issue of endogeneity by applying treatment effects model and propensity score matching (PSM) for the participants and non-participants of microfinance programmes. It has been confirmed by treatment effects model applied to panel data that household access to loans for productive purposes from microfinance institutions (MFIs) significantly increased per capita household income, while simple household access to general loans from MFIs did not. This suggests that purpose and monitoring of how clients use the loans is important for increasing household income, and thus decreasing household poverty. However, the application of treatment effects model and PSM to each cross-sectional component of the panel data shows that the poverty reducing effect of MFI on poverty was significantly reduced over the years. This suggests the need of more attention to be drawn to the primary purpose of microcredit, that is, poverty reduction, and also to monitoring loan usages in the situations where the profits of MFIs became increasingly squeezed and their activities became more commercialised under severe competitions among MFIs in recent years.