When calculating the optimal portfolios, the traditional mean-risk (including variance, value-at-risk (VaR), conditional value at risk (CVaR)) optimization model often assumes that mean returns are known constant values. In actual asset allocation, however, estimation of mean return will have deviation, namely there exists risk of estimation. On the basis of estimating the risk measured by CVaR, this paper further studies CVaR robust mean-CVaR portfolio optimization model and presents two different optimization algorithms, namely, the dual method and the smoothing method. Moreover, we explore some properties and characteristics of the two methods. Finally, we give some numerical experiments to show the feasibility and effectiveness of these two methods.