The ERFI survey (Étude des relations familiales et intergénérationnelles) is the French version of the international Generations and Gender Survey. The same respondents were interviewed three times, in 2005, 2008 and 2011. Although the survey was designed to avoid redundant questions from one wave to the next, a respondent’s family situation is likely to change over time. In each wave respondents were therefore asked about their children living in the dwelling, those living elsewhere and those deceased. The results of these separate questionnaire modules were then summed to give the total number of children (no direct question about the total was included). Substantial under-reporting of children in Waves 2 and 3 of the French survey was observed. This article aims to measure the extent of the omissions in order to alert potential users of the data, to identify which children “disappear” and establish whether this corresponds to particular respondent profiles. More broadly, the study calls into question the method of summing the children recorded in various parts of the same questionnaire in order to determine the respondent’s total number of children, and suggests that similar verifications should be made in other GGS surveys.