Returns to education remain an important parameter of interest in economic analysis. A large literature estimates returns to education in the labor market, often carefully addressing issues such as selection, into wage employment and in terms of completed schooling. There has been much less exploration of whether estimated returns are robust to survey design. Specifically, do returns to education differ depending on how information about wage work is collected? Using a survey experiment in Tanzania, this paper investigates whether survey methods matter for estimating mincerian returns to education. The results show that estimated returns vary by questionnaire design, but not by whether the information on employment and wages is self-reported or collected by a proxy respondent (another household member). The differences due to questionnaire type are substantial varying from 6 percentage points higher returns to education for the highest educated men, to 14 percentage points higher for the least educated women, after allowing for non-linearity and endogeneity in the estimation of these parameters. These differences are of similar magnitudes as the bias in OLS estimation, which receives considerable attention in the literature. The findings underline that survey design matters for the estimation of structural parameters, and that care is needed when comparing across contexts and over time, in particular when data is generated by different surveys.