Fault feature extraction is the key for fault diagnosis, and automatic feature extraction has been a hot topic recently. Deep learning is a breakthrough in artificial intelligence and known as an automatic feature learning method. One of the most important aspects to measure the extracted features is sparsity. Thus, this paper presents a stacked sparse autoencoder (SAE)-based method for automatic feature extraction and fault diagnosis of rotating machinery. The penalty term of the SAE can help automatically extract sparse and representative features. Experiments and comparisons are conducted to validate the effectiveness and superiority of the proposed method. Results fully demonstrate that the stacked SAE-based diagnosis method can automatically extract more representative high-level features and perform better than the traditional intelligent fault diagnosis method like artificial neural network (ANN).