In order to retrospectively assess exposure to agricultural pesticides in epidemiological studies, an indicator of agricultural exposure over past periods in France was developed using a semi-automatic semantic segmentation tool, and integrating wind direction. We evaluated the performance of this indicator by applying it to an epidemiological study. The land cover (LC) around the birth addresses of the 1,155 subjects of the national TESTIS case-control study was characterized from black-and-white photo-aerial images processed with the GOURAMIC semantic segmentation tool. A sample of test images was used to assess inter-operator photo-interpretation reliability among 14 novice operators compared to one expert operator. Images processed by the GOURAMIC software were compared to two LC databases over recent periods. Finally, wind data from MétéoFrance were integrated in the calculation of agricultural exposure. The LC around the households of TESTIS controls was consistent with the French epidemio- logic literature. Assessment of inter-operator variation showed moderate to strong agreement (kappa coefficients 0.49 to 0.71). Comparison of GOURAMIC and Corine Land Cover data- base showed agreement ranging from 20% to 86% depending on the land cover type. Incor- porating wind direction did not change the exposure level of the subjects. The proposed agricultural exposure indicator is relevant for epidemiological studies of chronic diseases in France. The study underlines the importance of training before using the GOURAMIC tool.