Abstract: Software reliability prediction classifies software modules as fault-prone module and not fault-prone module at the early age of software development.This paper proposes an improved support vector machine to predict software reliability.As a solution to the difficulty of choosing parameters，it applies Genetic Algorithm（GA） to choose the parameters of Support Vector Machines（SVM），and puts forward a model of predicting the software reliability based on GA-SVM.The method of principal component analysis is conducted to reduce the dimension of software metrics data.The simulation results show that this model can achieve the more precise prediction result than the prediction models of SVM，BP neural network，CART and clustering.