The aim of the research contained in this thesis was to develop a model of long-term upland catchment runoff that can be used for ungauged catchments. This is a problem due to the complex spatial and temporal nature of runoff and the main contributing processes, precipitation (P) and evapotranspiration (Et). It is also a problem due to the lack of suitable data on which to base and test models of these processes, particularly in remote upland areas such as the north-west of England, the study area of this research. Long-term runoff is important since it represents the maximum rate at which water is available for human use and management, for assessment of water resource yield and for prediction of extreme events that are particularly important in respect to climate change. Methods currently in use by water companies in the UK, such as North West Water Limited (NWW), are inadequate since they fail to account for the spatial and temporal nature of runoff. New more reliable methods are therefore required which will equip water managers with flexible and responsive runoff modelling tools based upon routinely available data and that are sensitive to the complex physical nature of the processes involved. A physically based distributed runoff model was developed using GIS technology and spatial data to interpolate and extrapolate available point-based hydrometeorological data. The strategy required the development of models to derive areal representations of P and Et. For the P modelling several interpolation techniques and artificial neural network models were investigated. The results were evaluated against an independent data set. The results showed that a geostatistical interpolation technique, detrended Kriging, which uses pointbased precipitation and spatial elevation data provided the most accurate estimates when compared to other methods. The models of Et involved extrapolation of point-based Et values derived from the Penman-Monteith formula (Monteith, 1965), using spatial land cover data. A point-based temperature function model (Wright and Harding, 1993) that reduces the Penman estimates of Et for upland sites was spatially implemented using spatial temperature and elevation data. No independent data were available for model evaluation but first estimates of errors were gained through comparison of errors of runoff and precipitation estimates. Overall it was found that the most accurate E, model results were derived when the temperature function model was not implemented. Evidence of whether or not a lumped or heterogeneous land cover representation provided the more accurate results was unclear. Error evaluation and sensitivity analysis of the modelled runoff was carried out using measured runoff records and the results were compared to those produced using the North West Water model. It was found that the GIS-based model provided improved estimates of long-term average annual runoff for upland catchments. The largest component of the errors of the GIS-based method were associated with the Et estimates. This was principally a result of poor quality and limited availability of data for the study area. The research highlights many wider issues related to the use of GIS and spatial data for hydrological modelling.