The performance of the STICS soil-crop model for the dynamic prediction of soil water content (SWC) and soil mineral nitrogen (SMN) in the root zone (120 cm) of seven agricultural fields was evaluated using field measurements in a coarse-grained alluvial aquifer of the Garonne River floodplain (southwestern France) from 2005 to 2007. The STICS model was used to simulate drainage and nitrate concentration in drainage water in all the agricultural fields of the study area, in order to quantify and assess the temporal and spatial variability of nitrate leaching into groundwater. Simulations of SWC and SMN in the seven monitored fields were found to be satisfactory as indicated by root mean square error (RMSE) and model efficiency being 6.8 and 0.84% for SWC and 22.8 and 0.92% for SMN, respectively. On average, SWC was slightly overestimated by a mean difference of 10 mm (3%) and there was almost no bias in SMN estimations (<0.5%). These satisfactory results demonstrate the potential for using the STICS model to accurately simulate nitrate leaching. Across the study area, simulated drainage and nitrate concentration were extremely variable from one field to another. For some fields, simulated mean annual nitrate concentration in drainage water exceeded 300 mg NO3 − L−1 and predicted nitrate leaching was close to 100 kg N ha−1, while other fields had very low nitrate losses. About 15% of the farmers’ fields were responsible for 60–70% of nitrate leaching. The SMN in late autumn, before winter drainage, was found the main determining factor explaining this variability. This situation may be attributed to unsatisfactory cumulative nitrogen management over the medium term. Ineffective nitrogen management was found to be more detrimental than a single annualincident of overfertilization, particularly in situations of deep soils and in cases of low or highly variable drainage between years.