A depth estimation algorithm for monocular infrared image based on nonlinear learning model is proposed in this paper. Firstly, some infrared image features corresponding to the depth are extracted using stepwise regression and independent component analysis. Secondly, we regress and train the features and the corresponding depth map of an infrared image based on BP neural network. A nonlinear depth estimation model is obtained. We can estimate depth distribution of an infrared image using the model. The experimental results show that more accurate depth information can obtained based on the proposed model comparing the linear estimation model.