In part I of this study (Xu et al., 2011), The ensemble-based 3DVar (three-dimensional variational technique) method with SVD (singular value decomposition) technique (SVD-En3DVar) is used for assimilation of the simulated radar velocity data and the results demonstrate that using the initial perturbation samples produced with 3DVar method in SVD-En3DVar can shorten the time interval of assimilation cycle and improve the short-term forecast of precipitation. In the current study the feasibility of using SVD-En3DVar for assimilating radar velocity observations is tested with the real observational data. Two torrential rain cases (June 2008 in South China and July 2003 in the Changjiang－Huaihe region) are chosen for the test and the 18-hour forecast of rainfall is compared with that by WRF-3DVar (3DVar based on the weather research forecasting model) assimilation. For the first case (2008) the observational data from 13 radars are assimilated and the forecast of rainfall within 18 hours is improved after assimilation with SVD-En3DVar, but the improvement is not evident with WRF-3DVar assimilation. For the second case (2003), only single-radar observations are used and the forecast of rainfall is improved in the first 6 hours after assimilation with SVD-En3DVar, however the forecasts are not improved by using either SVD-En3DVar or WRF-3DVar in the subsequent 12 hours.