Thermal imaging can reflect the temperature distributions of the scenes. The depth estimation of thermal imaging can reconstruct 3D temperature field of the scene,so it is significant in fault diagnosis and night vision navigation. A non-parametric depth sampling method for depth estimation for monocular thermal image is proposed. To overcome the lack of textures and blurry contours,a spatial pyramid matching method was used to analyze the features of thermal images. Firstly,based on SPM feature matching method,candidate images that have similar scene to the image of estimating depth were selected from the database. Then,a SIFT Flow algorithm was used to sample the depth maps of the candidate thermal images,and the depth information was transferred to the image of estimating depth. The experimental results show that this method can estimate the depth of monocular thermal image effectively,and it is better than other similar algorithms.