The annualized agricultural non-point source pollution model (AnnAGNPS) was applied to simulate losses of triazine herbicides to the River Cauca following application to sugarcane, maize and sorghum in the Cauca Valley of Colombia. Surface runoff was found to be the main driver of triazine losses to surface water in the catchment. Satisfactory simulation and validation of the hydrology was achieved after little calibration (Nash-Sutcliffe model efficiency = 0.70 and r2 = 0.73). A fairly good simulation of pesticides was generally achieved, but some patterns in the measured data could not be simulated. Uncertainty analyses of sensitive input parameters were carried out which explained most of the concentrations that were not captured by the initial simulation; however, evidence of point source pollution was observed for some large concentrations measured upstream. Replacing triazine herbicides with mesotrione was predicted to result in an 87% reduction in pesticide losses expressed as a proportion of the total pesticide applied.