This paper investigates the impact of Support Vector Machine Recursive Feature Elimination SVM-RFE method used for feature selection and feature ranking on speaker recognition performance in network environment. The motivation behind reducing the dimension of the feature set is by the fact that features are not all equally important to identify a speakere. In the present work, we thought to use SVM-RFE based feature selection to remove the irrelevant features influenced by speech coding algorithms, transmission errors and environmental noise of decoded speech. We find that the SVM-RFE selection method achieves comparable performance on network speaker recognition (NSR) system, while it obtains excellent performance with only few features. Result demonstrate the effectiveness of the feature selection methos on the transcoded TIMIT databse obtained using G722.2 speech coder together with the 6.60Kbit/s, 8.85Kbit/s, 12.65Kbit/s and 23.85Kbit/s bit-rates.