In this paper, we propose a hybrid method for intrusion detection which is based on k-means, naive-bayes and back propagation neural network (KBB). Initially we apply k-means which is partition-based, unsupervised cluster analysis method. In the form of clusters, we attain the gathered data which can be easily processed and learned by any machine learning algorithm. These outcomes are provided to the bayesian classifier which is a supervised learning method based on probability model. Fit and essential data attributes are obtained during this. Through filtered data learning is performed by back propagation neural network which is able to learn the patterns with less number of training cycles. We use KDD cup99's dataset. Through the bayesian classifier we detect attacks as DoS, U2R, R2L and probe. In this paper the main focus is given over classification and performance. Therefore different classification algorithms are applied for filtering the data set features.