The cloud base height is an important macrophysical parameter. Based on the data of visible, infrared and millimeter waves detected by MODIS and CPR, this paper proposes a new method to retrieve cloud base height by using Principal Component Analysis (PCA) and BP neural network. The cloud base heights retrieved by PCA-BP method and CloudSat products are compared with the cloud base height obtained from the sounding data processed by the relative humidity threshold method. The results show that the value of deviation of PCA-BP method is lower than that of CloudSat products for most types of clouds. The cloud base heights retrieved by PCA-BP method and CloudSat products have lower values in summer and higher values in other seasons, and the root mean square errors occurred by using the PCA-BP method and the radiosonde have lower values than those occurred by using the CloudSat products in all seasons, the cloud base heights retrieved by the two have a consistent seasonal characteristic, that is, the values are the highest in summer and the lowest in winter. The cloud base heights retrieved by PCA-BP method and CloudSat products decrease with increasing latitude. The two method have different retrieving results in different regions, the biases of both two methods decrease with increasing latitude. The comparison results show that it is to some extent feasible to use the PCA-BP method to retrieve the cloud base height.