The French National Health Insurance’s drug reimbursement databases offer an opportunity for epidemiological surveillance by enabling the identification of patients with certain diseases based on their medication use. Before considering their use for this purpose, it is necessary to assess their validity. We studied their validity for identifying patients treated for Parkinson’s disease in five branches of the Mutualité sociale agricole in 2007. We compared the patterns of antiparkinsonian medication use among patients whose diagnosis of Parkinson’s disease was confirmed by a neurologist and among patients treated for another condition. We developed various predictive models for Parkinson’s disease to estimate the probability that users of antiparkinsonian medications are being treated for this condition. Among 1,114 participants who used an antiparkinsonian medication in 2007, a diagnosis of Parkinson’s disease was confirmed for 320 (29%) of them, while 794 (71%) were treated for another reason. Various models were compared in terms of prediction quality. The logistic model, which includes cumulative doses of different types of antiparkinsonian drugs and treatment regularity, performed well (C-statistic = 0.953; sensitivity = 92.5%; specificity = 86.4%). According to this model, the prevalence of Parkinson’s disease among individuals aged 18 and older is 6.66 per 1,000; adjusting for the model’s sensitivity and specificity yields a similar estimate (6.01 per 1,000). We also present an example of a study examining the relationship between the prevalence estimated in this way and exposure defined at the geographic level. We discuss the advantages and disadvantages of drug reimbursement databases for monitoring the incidence of certain diseases and studying their risk factors. Our results show that, for certain conditions, a surveillance system based on drug reimbursement databases—without the need to collect additional information from individuals—could be implemented at the national level. (R.A.)