Objective - In France, cancer registries cover only a portion of the departments. To achieve comprehensive coverage in terms of morbidity, the use of medical-administrative databases (MDBs) has been proposed. The objective of this study was to examine the characteristics of BDMA necessary to enable their use not only as a tool for epidemiological monitoring of cancers, but also to identify cases within the context of cohort follow-up or to conduct surveillance outside an area covered by a registry. Method - We analyzed the concordance, at the individual level, between data on "long-term conditions" (ALD) on the one hand, data from the Medical Information Systems Program (PMSI) on the other, and data from departmental cancer registries, used in this study as the reference for incidence. This study focused on cases diagnosed in 2004. Results - For the 2004 ALD data, the 5 registries included in the study recorded 22,455 cancers, which were compared to 17,133 ALD cases. For the 2004 PMSI, the 4 registries included in the study recorded 16,066 cancers, which were compared to 26,294 cancer diagnoses provided by the Medical Information Departments of 68 institutions. The results show that for identifying new cases of cancer, the sensitivity of ALDs (0.56) is lower than that of the PMSI (0.70). However, the positive predictive value (PPV) of an ALD report is higher than that of a PMSI report (0.74 versus 0.43), as the PMSI includes many more prevalent cases than the ALDs. These results vary depending on the cancers studied, as there are multiple factors contributing to underreporting. Conclusion - ALDs appear difficult to use alone to conduct a comprehensive case census. However, they are a much better indicator of whether a person actually has a prevalent cancer. The PMSI appears to be more comprehensive, but its use would require screening many cases due to its low PPV. More complex record extraction criteria than those we used, capable of distinguishing incident cases from prevalent ones, are necessary. However, overly stringent selection criteria risk compromising the PMSI databases’ ability to identify cases. (R.A.)