The uncertainty of water quality predictions caused by uncertainty in the inputs related to emissions of diffuse pollution is analysed. An uncertainty analysis of the effects of diffuse pollution is essential to compare the cost and benefits of measures to lower those emissions. We focus on diffuse nitrate pollution due to fertiliser use. Using an efficient Monte Carlo method and Latin Hypercube sampling, the contribution to the overall uncertainty by each of the inputs is calculated. The modelling environment is ESWAT, an extension of SWAT, which allows for integral modelling of the water quantity and quality processes in river basins. The diffuse pollution sources are assessed by considering crop and soil processes. The crop simulations include growth, uptake of water and nutrients and several land management practices. The in-stream water quality model is based on QUAL2E. The spatial variability of the terrain strongly affects the non-point source pollution processes. The methodology is applied to the Dender basin in Belgium. Eight inputs have significant influence on the time that the nitrate content in the river is higher than 3 mg/l. The uncertainty analysis indicated wide uncertainty bounds (95% percentile bounds differ up to +/-50% from the average NO3 predictions).