Varied data mining techniques have been developed for intrusion detection. However, it is unclear which data mining technique is most effective. In this paper, we present our research work in developing a Multi-Class SLIPPER (MC-SLIPPER) system for intrusion detection to learn whether we can get benefit from boosting based learning algorithm. The key idea is to use the available binary SLIPPER as a basic module, which is a rule learner based on confidence-rated boosting. Multiple arbitral strategies based on prediction confidence are proposed to arbitrate results from all binary SLIPPER modules. Our system is evaluated on the KDDCUP?99 intrusion detection dataset. The experimental results show that we get best performance using a 5-7-5 BP neural network; and the performance using other arbitral strategies are better than the winner of the contest does in term of misclassification cost (MC).