Urban drainage models are examined in terms of model purpose, type and usage. Model usage for the prediction of flooding behaviour in the UK adopts the design storm approach, which assumes that flood frequency equals rainfall frequency. This equation is shown to be affected by antecedent moisture conditions and rainfall spatial variability. Data collected from three small urban catchments (0.3 ha, 0.7 ha and 3.2 ha) in Southampton are used to calibrate and verify three urban drainage models: namely WASSP-SIM, WALLRUS-SIM and USGS-DR3M. A probabilistic approach to model verification is used and demonstrates that accepted bounds to model accuracy associated with the use of these models are difficult to justify if consideration is given to the possible errors in rainfall and discharge measurement. Comparison of the results from storms collected during 1985 on these catchments shows that the DR3M performs better than either of the other models in terms of a number of standard indices of model accuracy. Soil moisture observations collected over a two year period using the neutron probe method are then used to calibrate and test a number of simple soil moisture models including the soil moisture algorithm incorporated in the MORECS procedure and that included in the DR3M. These models are found to predict catchment average soil moisture conditions to within 20 mm, using local rainfall information and standard evapotranspiration curves as inputs. Using the MORECS soil moisture model it is shown that antecedent soil moisture conditions do have an impact on observed percentage runoff values, but that the relationship is confused by seasonal factors. The soil moisture model is then used together with WASSP-SIM to derive discharge-frequency curves using a continuous simulation approach. Comparisons with the design storm approach confirm the assumptions of the approach, although they do not prove that either approach produces the correct answers.