The deep learning method has been widely applied to the SAR target classification issue in recent years, which achieves a high classification accuracy compared with the conventional techniques [1]. Especially for the SAR image, the convolutional neural network (CNN) is always the first choice for its powerful capability of extracting features and generalization in the images. Researchers have designed a diversity of CNNs to finish the task, which only utilizes the raw SAR image as the input to the network. To train an active CNN, they typically conduct one step called data augmentation to enlarge the size of the original dataset. With more images to train the network, the much better classification accuracy they will get in the end. In this paper, we construct a 3-D convolutional neural network for the target classification in SAR images. Besides this, we also propose a new structure of the input to the 3-D CNN for the SAR image, which consists of three components of the raw SAR image. We only use the original size of the SAR image dataset to train our proposed 3-D CNN. Experiment results on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset show that our method achieves an average classification accuracy of 99.75% in the test set, which is superior to most of the state-of-the-art techniques so far.