The overall objective of this thesis is to better understand and assess pesticide leaching at the regional scale, using both the analysis of monitoring data and spatially distributed modelling. Atrazine contamination of the Brusselian aquifer (central Belgium) is poorly understood. Considerable uncertainty surrounds whether the pollution is agricultural or non-agricultural in origin. The spatial and temporal covariance of atrazine concentrations was studied by fitting semivariogram models to monitoring data. Correlation ranges were found to be 600 metres and 600-700 days. A non-parametric one-way ANOVA found a strong relationship between mean concentrations and land use, whilst other environmental variables were found to be less important. Higher levels of pollution were detected in areas dominated by urban land use suggesting that atrazine residues in groundwater resulted from non-agricultural applications. Modelling pesticide leaching at the regional scale (Dyle catchment) was used to assess groundwater vulnerability. Different approaches to process soil information were tested with both a linear (modified Attenuation Factor) and a non-linear (GeoPEARL) leaching model. The CI (calculate first, interpolate later) and IC (interpolate first, calculate later) approaches were identical for the linear model, but differences in the amount of leaching were found for the non-linear model. The CI approach would be expected to give better results than IC, but the CA (calculate alone) approach is probably the best method if no spatial output is required. Finally, a methodology was developed to quantify the uncertainty arising from the spatial variability of non-georeferenced parameters (i.e. those assumed to be spatially constant in deterministic simulations). A Monte Carlo analysis of atrazine leaching was performed with six pesticide and soil properties as uncertain inputs. Spatial variability of non-georeferenced parameters had a significant influence on the amount of simulated leaching. In the stochastic simulation, concentrations exist above the regulatory level of 0.1 µg/L, while virtually no leaching occurred in the deterministic simulation. Including the spatial variability of substance parameters (half-life, sorption coefficient...) would have significant consequences for future registration policies, especially if risk assessments are implemented in a spatially distributed way.