Facial paralysis is a common clinical condition occurring in the rate 20 to 25 patients per 100,000 people per year. An objective quantitative tool to support medical diagnostics is necessary. This paper proposes a robust method that decomposes the images into multi frequencies-space domain, and then features are extracted for classification using a support vector machine (SVM). The method analyses the images in frequency-space domain, so it overcomes the problems of other techniques such as the change of illumination, noise and redundant frequencies. Experiments show that our proposed method outperforms other techniques testing on a dynamic facial expression image database.