The paper first applies nonparametric kernel estimation method to estimating CVaR which is currently a popular risk measurement tool，then derives a two-step kernel estimator of CVaR with distribution-free specification． Next，a two-step kernel estimator of CVaR is embed into the mean-CVaR portfolio optimization models to derive financial risk estimation and portfolio optimization at the same time． A simple iterative algorithm is designed to solve these models． Monte Carlo simulation result shows that the portfolio optimization models and the algorithm based on the two-step kernel estimator of CVaＲ is feasible and effective，and that the estimated error of portfolio frontier is very small． The models and algorithm above apply to a risk-free security．Finally，an empirical analysis of daily return data from Chinese A-stock market is presented to illustrate the application of this research．