Abstract: In order to improve support vector regression (SVR) learning ability and generalization performance, a joint optimization method for selecting features and SVR parameters is proposed. By using the method, a new set of uncorrelated features is obtained by using principal component analysis , the mean squared error(MSE) is taken into account for the accuracy evaluation of SVR, and a real-coding based immune genetic algorithm is employed to solve the joint optimization problem. Simulation experiments show that the joint optimization method guarantees better regression accuracy and the optimization process has a higher rate than the single optimization of features or SVR parameters.