This article considers whether survey and census data offer comparable bases for estimating trends in women’s age at marriage in Africa. It uses the indicator of median age at first marriage calculated from the proportion of never-married women by age. It draws upon two bodies of data : first, a pan-African nuptiality database is used to assess differences between estimates drawn from the two types of source at the scale of the whole continent (453 censuses and national surveys undertaken since 1950 in the 55 countries of Africa) ; and second, data from 15 MICS surveys which record marital status twice (each respondent is included on both a household and an individual questionnaire) are analysed to pinpoint inconsistencies. The median age at first marriage is generally higher when estimated from census data than from survey data. Several error mechanisms combine to create this effect. In censuses, imprecise recording of marital status leads to overestimation of numbers never-married, and therefore to overestimation of median age at marriage. In surveys, meanwhile, the tendency to underestimate young women’s age, thereby excluding a disproportionate number from the survey sample of women aged 15-49, and the less thorough coverage of never-married women lead to under-representation of those never-married and therefore to underestimation of age at marriage. This analysis does not suggest that one type of source should be preferred over the other, but rather that neither source should be neglected.