This paper presents the novel feature selection method that maximizes class seperability between normal and attack patters of computer network connections. Recent years have witnessed increased interest in using a genetic algorithm to improve the performance of a classijer. In this paper we focus on selecting a robust feature subset based on the genetic optimization procedure in order to improve a true positive intrusion detection rate. During the evaluation phase, the performance of proposed approach is contrasted against one of state-of-the-art feature selection method using a naiie Bayesian class$er. Experimental results show that the proposed approach is especially efective in terms of detecting totally unknown attack patterns.