An experiment is designed to explore risk conception and evaluation of investors. A new risk measurement, which is called martingale variance, is included. The new measure uses previous return as benchmark instead of mean return to calculate variance. The results show that respondents are more care about expected loss than variance, but are indifferent between variance and loss probability. Meanwhile, the respondents prefer martingale semivariance to martingale variance significantly. The weighted martingale semivariance also dominates martingale semivariance. The results can be helpful to explain the relationship between risk and return.