In gun bore flaw detection based on machine vision, computer can not recognize flaw image with typical idea, such as target segmentation, feature extraction, feature analysis and judgment. Aiming at this problem, twice-spectrum analysis method was adopted to extract flaw information from overall features of image. Image twice-spectrum was defined as logarithmic amplitude spectrum of power spectrum. And four parameters, which were Length-to-Width Ratio of Bright Straight Line, Energy Mean, Energy Variance and Energy Moment, were extracted to analyze gun bore images. The experimental results testify rationality of the parameters.