This paper evaluates the denoising effect of NMF with different auxiliary variables on face images. Firstly, the data is preprocessed, and then different degrees of Gaussian noise or Salt-pepper noise is added, respectively. Finally, the denoising effects of NMF, L 1 -NMF, L 2,1 -NMF, CIM-NMF, and Huber-NMF are evaluated by three evaluation methods. The results show that all NMFs show good robustness in dealing with Gaussian noise; however, it is not ideal for dealing with Salt-pepper noise, and CIM-NMF is better than other NMFs in dealing with Salt-pepper noise.