The rapid development of Geographical Information Systems (GIS) has together with the inherent spatial nature of hydrological modelling led to an equally rapid development in the integration between GIS and hydrological models. The advantages of integration are particularly apparent in flood extent modelling. In this thesis, the integration of hydrological models and GIS is approached on the basis of performance, with performance taken as the balance of computational efficiency, flexibility of application, and most importantly the reliability of the integrated model. It is shown that predictive reliability is dominated by model uncertainties, particularly in model roughness parameters. These roughness parameters are found to be more conceptual than physical as they represent bulk momentum loss parameters at the reach scale. Limited data on spatial extent of flooding is available to constrain these uncertainties, and where such data is lacking the simplest numerical approach may be as reliable as more complex approaches. The overall performance of the simple approach is then higher as this is more easily integrated within GIS. Observations of flood extent from aerial photographs may help constrain uncertainties, though much more value is found from distributed water level observations in the floodplain. The lack of hydrological data also results in high resolution GIS data of elevation or land use being of limited value. As sufficient hydrological data is unavailable and perhaps impossible to acquire, model predictions made are recommended to be considered probabilistically, irrespective the level of integration with GIS.