The prediction of severe convection is a major forecasting problem in Australia during the summer months. In particular, severe convection in the Sydney basin frequently produces heavy rain or hail, flash flooding, and destructive winds. Convective activity is a forecasting challenge for the Sydney basin, mainly from October to April. Currently, there is a need for improved numerical model guidance to supplement the official probabilistic convective outlooks, issued by the operational forecasters. In this study we assess the performance of a very high resolution (2 km) numerical weather prediction (NWP) model in terms of how well it performed in providing guidance on heavy rainfall and hail, as well as other mesoscale features such as low level convergence lines. Two cases are described in which the operational forecasts were incorrect on both occasions. Non-severe thunderstorms were predicted on 1 December 2000 but severe convection occurred. Severe convection was predicted on 8 December 2000, but no convection was reported. In contrast, the numerical model performed well, accurately predicting severe convection on 1 December and no convection on 8 December. These results have encouraged a program aimed at providing an enhanced numerical modelling capability to the operational forecasters for the Sydney basin.