For the combined spacecraft consisting of the servicing spacecraft and noncooperative target, the measured angular rate signals of the combined spacecraft are often affected by large noises, which will influence the subsequent control accuracy seriously. Due to the unknown inertia tensor of the noncooperative target, the signal cannot be effectively denoised using model information. To solve the abovementioned problem, an angular rate signals denoising method of the combined spacecraft based on the deep learning method is proposed in this paper, which do not require the model information in advance, and is absolutely datadriven. Based on the deep learning, the generative process of the training data which is needed for angular rate denoising is firstly presented, and then a deep network model for angular rate signals denoising of the combined spacecraft is constructed, including the methods of optimization and parameter initialization of the model. Finally, denoising effects are compared between the proposed method and the wavelet denoising method using test data. The result shows that the proposed deep learning method has a better denoising effect than the wavelet method.