Aiming to solve the problem that bilateral filter preserves noise boundaries which have large intensity difference as strong edges, Bilateral Tensor Filter (BTF) was presented. Based on bilateral filter, the structure preserving function was constructed self-adaptively by local structure anisotropic similarity and photometric similarity, which filter the image along the direction of the local structure around the edge and turn into approximate isotropic around the flat area in the presence of strong noise. By substituting this structure preserving function for edge preserving function, the structure information was preserved while the noise was suppressed. Experiment results demonstrate that our bilateral tensor filtering algorithm substantially improves the quality of the filtered images and possesses a stronger edge preserving ability.