Child labor statistics are critical for assessing the extent and nature of child labor activities in developing countries. In practice, widespread variation exists in how child labor is measured. Questionnaire modules vary across countries and within countries over time along several dimensions, including respondent type and the structure of the questionnaire. Little is known about the effect of these differences on child labor statistics. This paper presents the results from a randomized survey experiment in Tanzania focusing on two survey design choices: different questionnaire design to classify children work and proxy response versus self-reporting. Use of a short module compared with a more detailed questionnaire has a statistically significant effect, especially on child labor force participation rates, and, to a lesser extent, on working hours. Proxy reports do not differ significantly from a child's self-report. Further analysis demonstrates that survey design choices affect the coefficient estimates of some determinants of child labor in a child labor supply equation. The results suggest that low-cost changes to questionnaire design will potentially clarify the concept of work for respondents.