According to the characteristics of infrared imaging and the gradual change of intensity levels of metal surface defects, a vision inspection method for surface defects of metal based on statistically analyzing wavelet texture has been proposed. Firstly, the CCD sensors are used to obtain infrared video-data for surface of copper strips, and then the first-order Haar wavelet is used to decompose infrared image. Secondly, two multivariate statistical methods, including Hotelling T2 control chart and Chi square test, are used to fuse the four wavelet characteristics. Finally, the statistical values are used to distinguish the existence of defects and classify the defects using support vector machine. The capabilities of two kinds of wavelet-domain-based multivariate statistical approaches in inspecting defects have been researched deeply. The experimental results demonstrate that the Hotelling T2 method gets the better performance, which achieves a 92.8% probability of detecting the existence of micro defects and a 95.42% probability of classifying the defects.