For the situation of three or more diagnostic results in a diagnostic test， we often use ROC surface instead of ROC curve to judge the accuracy of the test. Reviewing the previous literature on ROC surface estimation， we can find that many scholars use parametric methods and non-parametric methods， and the results obtained by using semi-parametric methods are also excellent. Combining continuation ratio logistic regression model and bootstrap method, it can be found that the ROC surface obtained is smoother and more accurate than that of the non-parametric method. When calculating the semi-parametric maximum likelihood estimation， the semi-parametric method can use some logistic programs in some statistical software to replace the Newton’s method, which can quickly get results. Therefore， the semi-parametric method is obviously better than the non-parametric method in the ROC surface estimation problem.