The study addresses the comparability of child labour estimates produced by different common household survey instruments. This question has important implications for credibility of published estimates of child labour, and for the reliability of current survey instruments as tools for monitoring and guiding efforts towards the progressive elimination of child labour. The study, covering some 87 datasets for 35 countries, first confirms that estimates of child labour vary significantly across different kinds of surveys. The variation, moreover, appears to be substantially larger than that relative to other children’s activities like schooling. The study then addresses whether the observed significant differences in estimates are due to difference in population characteristics or to other features of the surveys. In other words, whether different populations are targeted by the various surveys, or whether they address the same (or very similar) population with different instruments. The empirical results indicate that it is the latter explanation, i.e., differences in survey features are most relevant. Differences in observable survey characteristics such as questionnaire type and fieldwork season explain some of the variation in child labour estimates across survey instruments, but a larger part of the variation stems from unobservable survey characteristics. Elements of the survey process not spelled out in the survey documentation, such as interview methods, the familiarity of interviewers with child labour concepts, the accuracy of data coding and processing, are all likely to be important in this context.