We use a simple multiplicative spatio-temporal model to describe variability in a sequence of water quality monitoring data from headwater streams in the Conwy catchment, North Wales. The spatial component of the model treats concentrations as due to simple mixing of a small number of distinct source types, each associated with particular upstream catchment characteristics. The temporal component allows concentration variability due to seasonal or hydrological change. We apply the model using three candidate catchment characteristic classifications to generate mixing concentrations, and a seasonal component to describe temporal variability, and test a range of sub-models. We identify a cross-classification of soil and land cover as providing the best spatial indicator of water quality of the classifications considered. The spatial model based on a selected grouped cross-classification was shown to account for between 35% and 90% of the spatial variability and the seasonal model accounted for between 45% and 100% of the temporal variability in the data. Analysis of residuals showed an inverse relationship between DOC and sulphate and between hydrogen ion concentration and calcium and magnesium. We also found residual correlations between sites which are strongly related to landscape class. These are attributed to landscape class by time interactions which are not accounted for in the simple multiplicative model.