OBJECTIVE： To provide methodological reference for converting non-utility scale measurement results into health utility values. METHODS： Referring to domestic and foreign literatures， mapping methods and relevant models in health utility measurement were summarized. The effect of each model on probability mapping was introduced by taking the Medical outcomes study 12-item short form health survey measurement results converting into the EuroQol group’s 5-domain utility values as example. RESULTS： The mapping methods can be adopted to obtain the health utility values by establishing the mapping relationship between non-utility scale and utility scale. The common models included ordinary least square （OLS） model，censored least absolute deviations （CLAD） model， Tobit model，multinomial Logistic regression （MNL） model，Bayesian networks （BN） model， etc. OLS model was relatively simple with a good predictive validity，but it would be limited by the ceiling effect； Tobit model was not limited by the ceiling or floor effect，when the error term satisfied the variance homogeneity and normality， prediction result of  Tobit model was better than OLS model； CLAD model can be used for the situation of Tobit model unsuitable for non-variance homogeneity of the error term； MNL model firstly determined a health state by regression analysis and then determined its utility value； the prediction validity of BN model was good and it didn’t involved many assumptions and restrictions condition in econometrics， but the construction of BN model was greatly influenced by domain experts. Main methods for the calculation of health utility value with MNL or BN model were Monte Carlo simulation method， expected-utility method， most-likely probability methed， etc. We can carry out the model performance evaluation by using the R2， the adjusted R2， the mean error， the mean squared error and the mean absolute error， and then select the optimal model to calculate health utility values. CONCLUSIONS： Due to the advantages and disadvantages of each mapping model， it is necessary to select different mapping models based on the actual conditions.