Epidemiologic studies on adverse health effects of cellular telephone use have assessed exposure either by self-reported use based on questionnaire data or by using data on subscriptions for a cellular telephone provided by network operators. With the latter approach, subjects are misclassified when they regularly use a cellular telephone subscribed in someone else's or in a company name or when they subscribe for a cellular telephone which they use only occasionally. Self-reported use is hampered by recall difficulties and possible differential participation by exposure. In Denmark, we conducted a retrospective cohort study of cellular telephone subscribers (including the entire Danish population) and a case-control study on brain tumors and cellular telephone use (with 1355 participants) and, thus, had the opportunity to compare the two exposure measures with two large-scale data sets, using self-reported use as a "gold standard." Overall, there was a fair agreement (kappa value of 0.30, 95% confidence interval 0.23-0.36), with a low sensitivity (30%) and a high specificity (94%). Agreement was slightly better for controls, and low-grade glioma cases compared to high-grade glioma cases and meningioma cases. A comparison of odds ratios (OR) of the case-control data set based on either self-reported use or on subscriber data shows no major differences, giving OR of 0.7 and 0.6 for acoustic neuroma, 0.9 and 1.1 for glioma and 0.9 and 0.7 for meningioma. A discussion of the two exposure measures reveals that they both have limitations with regard to a potential underestimation of an association and there is some concern whether they are good enough to allow a detection of possibly only subtle changes in risk. These limitations can be minimized in prospective follow-up studies.