This paper aims to explore the discrepancy between self-reported and test-based literacy estimates in Bangladesh. It uses the Education Watch national literacy survey data of 2002, during which the literacy status of a nationally representative sample was identified using the two methods referred to above. The findings revealed that the literacy rate generated through a literacy test was significantly lower than that found when the self-report method was used. The level of discrepancy varied from one population group to another, indicating that literacy assessment using the self-report method cannot be equally appropriate for all population groups. Those with a few years of schooling were at a particular risk of over-estimating their status. The paper makes a case for a written literacy assessment rather than an oral evaluation.