Based on the CV-SVM principle and methods, the paper constructed the runoff time series model by combining phase space reconstruction theory with support vector machine. According to the many methods of phase space reconstruction to determine the delay time τ and embedding dimension m, and the shortcomings of determining different results by every method, this paper adopted the method of trial and error, the delay time τ and embedding dimension m value range from 2to 10in order to construct 81monthly runoff simulation model based on the theory of phase space reconstruction CV-SVM. Taking monthly runoff simulation in Nanli River Donghu Lake station for example, it determined the optimal delay time τ and embedding dimension m and compared the results with that determined by autocorrelation function. The aim is to explore the influence of delay time τ and embedding dimension m of monthly runoff simulation on accuracy and its regularity, and provide method and reference for the hydrological simulation and forecast of time series.