To improve the prediction performance of chaotic time series, a new method is proposed for parameters joint optimization of phase space reconstruction and support vector machine (SVM). The main idea of the joint optimization method is that the parameters from phase space reconstruction and SVM are designed jointly using uniform design firstly, and then the parameters are optimized jointly based on self-calling SVM. The results tested by chaotic time series indicate that the proposed method has more advantages than traditional methods, such as better prediction accuracy and lower computational complexity.