By using the GRAPES （Global/Regional Assimilation and Prediction Enhanced System） Single Column Model and the GCSS WG4 Case 3 data, the numerical experiments were conducted to evaluate the cumulus parameterization schemes, such as BMJ, SAS and KFeta. The results show that BMJ and KFeta can simulate the precipitation that consistents with the observation, while SAS is relatively poor. The potential temperature errors of the middle and upper troposphere simulated by BMJ are smaller, but cool bias of the lower troposphere is serious, meanwhile, it has stronger ability to transport water vapor upward which leads to the lower troposphere drier, middle and high troposphere wetter, so that the atmosphere adjusted by BMJ is too stable. KFeta has a trend to be cooler and wetter at lower troposphere and mainly produces cold bias, but the magnitude of lower troposphere is smaller compared to other schemes, at upper levels serious cool bias is caused by excessive overshoot. In order to solve the problem of KFeta excessive overshoot, this paper proposes a new modified scheme in which the vertical velocity equation is modified, and the modified scheme proposed by Anderson et al. also is examined. Numerical experiments show that both modified schemes can reduce high level cold bias effectively; In general, the new modified scheme is more satisfactory.