Surveys to measure the extent of illicit drug use by young people produce contrasting results depending on the data collection method employed. This article compares two surveys, one using a self-completed questionnaire in school, the other a telephone survey conducted at home. The former yields a systematically higher prevalence of drug use over the last year. These differences, which are then analysed for cannabis consumption, seem not to result from selection bias. When analysed for senior high school (lycée) pupils only, they are unchanged after controlling for age and sex. Similar differences are observed on other sensitive questions (truancy, occasional tobacco consumption, suicidal thoughts), suggesting that they are indeed a result of the data collection method. A logistic model is used to evaluate the effect of the collection method, after adjusting for a number of socio-demographic and educational indicators: relative to the telephone interview, the self-completed questionnaire in school increases by 1.7 the probability of reporting cannabis use during the previous year. A possible “group effect” with this form of collection remains to be controlled for, however.