The use of multi-source data, especially the fusion of optical and radar images, is a promising way for improving the level of interpretability of the remote sensing data which leads to a significant performance improvement. In most practical situations, there are two important challenges in the image classification methods: feature space generation and choosing the appropriate method for feature selection. This paper aims at reducing the time required for achieving the optimal features. To realize this objective, a new feature selection method is suggested based on the fusion of optical and radar images. In the proposed method, Minimal Redundancy-Maximal Relevance (MRMR) and Genetic Algorithms (GA) are combined and the optimal features are seeked to improve the classification accuracy based on the SAR and optic data. In doing so, at first the SAR data and the optical image are fused together using wavelet procedure. Afterward, several features are extracted from the fused image. In the next stage, the feature selection step is carried out based on the GA method and the combination of the MRMR and GA which is termed MRMR-GA algorithm. Lastly, the fused image is classified using the support vector machine classifier. For the performance analysis of the proposed approach, the TerraSAR and the Ikonos images which are acquired over Shiraz in Iran, are employed. The suggested method leads to the overall accuracy of 97.25 percent, which indicates the accuracy of the MRMR-GA method is 3% higher than that of the SVM classification with inclusion of the entire feature set. Moreover, the overall accuracy of the proposed approach and GA is approximately equal, while the performance of the MRMR-GA method is approximately 2.5 times faster than the GA. Therefore, the obtained results confirm the efficiency of the proposed method in feature selection for the purpose of the image classification.