To design a highly automatic and practical method for target discrimination in synthetic aperture radar images, we propose in this paper an improved scheme consisting of the framework and algorithms for target discrimination. Our main contribution in this scheme comprises four aspects. First, an integrative frame sequentially combining the algorithm based on feature extraction and the knowledge of target group has been presented. Second, three new features for target discrimination have been introduced. Third, a genetic algorithm-based featureselection algorithm has been presented. The results show that this algorithm can evaluate the goodness-of-feature better. Finally, to improve the accuracy of the discriminator, we have designed a weighted quadratic distance discriminator, which has been observed to improve the performance of target discrimination. We have analyzed the performance of the proposed scheme comprehensively and specifically using some measured data, and carried out comparisons of the existing algorithms. The results show that the proposed scheme could improve the application ability in target discrimination.