Cloud visualization of weather forecast data plays an important role in atmosphere analysis and weather prediction. However, due to the low resolution of weather forecast data and complicated geometry of clouds, it is a challenging problem to present realistic three dimensional clouds. In this paper, we propose a new framework to solve this problem. Firstly, data points are classified into three types of clouds: cumulus, stratus and cirrus. Then, an interpolation method is used to construct large-scale stratus. For cirrus, we apply a searching process. Finally, clouds are rendered in interactive way with various view directions. Experimental results demonstrate that the proposed method can yield large-scale and realistic clouds.