An Ensemble Square Root Filter (EnSRF) data assimilation system concerning storm scale issues is built under the Weather Research Forecast (WRF) model framework and is tested with different types of severe weather events to examine the performance of this data assimilation system in convective scale application. One heavy rainfall event (5 July 2003) and one strong convective event (5 June 2009) are selected for conducting the tests. Results show that the analysis ensemble means at 60 min in both cases and the 30 min forecasts from these analyses are close to the observations. The standard deviations of increment in observation space reach 3 m/s to 4 m/s and 9 dBz to 11 dBz for radial velocity and reflectivity respectively. The impacts of different initial perturbation strategies are investigated in this paper, namely, adding initial perturbations to the entire computational domain and to the limited region around the observed storm area, perturbations to the environmental wind field or not. Results show that perturbing the limited region does not improve the analysis results compared to perturbing the entire forecast domain, while perturbing the environment wind field produces better results in the Meiyu heavy rain case, indicating the more significant impacts of environment uncertainty on relatively larger scale weather events. The impacts of data assimilation on precipitation forecast are also examined. In the heavy rain case, the forecast results show that within 6 h, both the single forecast and the ensemble forecast initialized from analysis obtained through the assimilation of radar observations are better than the corresponding result in the experiment without data assimilation. The above-mentioned experiment results indicate that this data assimilation system has the ability to be applied in storm scale issues. 