We propose an adaptive surveillance video noise filter (ASVNF) using models for marginal distributions of wavelet coefficients. In order to suppress mixture Poisson-Gaussian noise for surveillance video, the wavelet domain based denoising function in the ASVNF adapts its output to the local spatial video structure and the property of the video noise. Based on the adaptability, the ASVNF recovers the original video signal from the noisy observation while it preserves the fine structure of the video. Experiments conducted using a wide range of test video sequences with different noise levels have demonstrated that the ASVNF is superior to a number of benchmark methods, in terms of objective measurements and visual image quality.