Abstract: Expansion of computer network causes the network to be the target of attacks. In this study, using the idea of mixture experts efficient and effective intrusion detection system is presented. In this method of KDD 99 data set for training and test data and using experts algorithms to reduce the dimension attributes that are less important in identifying attacks and eliminating the overhead of data input has reduced. In order to make different errors in the output of the three types of classification, support vector machines, decision trees and neural networks, multi-layer perceptron is used. Makes different errors, experts can identify weaknesses in the other attacks to different causes, including lack of training goes back, can identify. Also in this method of learning the weaknesses of each experts, to provide experts management is a possible mediator. experts mediator network to allow such actions in the output shows, that the new sample should decide otherwise with relevant experts to filter the output of the impact on the final result. The results of this study indicate optimal performance of this system is acceptable.