Two-factor Vasicek model in state-space framework with Kalman filtering is applied to research treasury term structure of interest rates in Shanghai Stock Exchange. The observation errors of 1-year and 20-year interest rates were extracted, estimated the marginal distribution with the nonparametric method under the condition of no-specific distribution assumption. It calibrates the Archimedean Copula and mix copula by MLE. We find that the Gumbel Copula and Mix Copula are more suitable. At last, the VaR of Treasury Portfolio is valued by a new Mont Carlo simulation approach , we conclude that it would underestimate the risk to assume that the dependence structure follow Gaussian Copula, Frank Copula, Clayton Copula and mix Copula, Gumbel Copula is more superior.