Noise type recognition for noise images is one of the key techniques for targeted denoising of noise images. Support vector machine (SVM) is a classification method based on statistical learning theory applicable to finite sample cases, and its classification ability depends largely on its related parameters.In this paper anew method is proposed to optimize the parameters of SVM. Grey wolf optimization (GWO) algorithm is used to optimize the parameter of SVM for obtain the optimal classification model，meanwhile the proposed method is applied to the noise type recognition experiment of noise images. The images withnoise interference are formed by three types of noises such as Gauss noise, Salt-and-Peppernoise and speckle noise. 90 sample data are taken as training samples and the remaining 60 sample data are taken as the testing samples. The Zernike moments and wavelet high-frequency non-significant coefficient subband energy ratio are selected as the eigenvalue. The GWO-SVM classifier is used to classify noise images. The experimental results show that the GWO-SVM method has better classification accuracy than the traditional SVM classifier.