Abstract: Snow depth is an important parameter in snow monitoring.How to select the charactersitic parameter and construct model is one of the key issues in snow depth retrieval using remotely sensed images.In this paper,MODIS images over Tianshan region were used,from which 37 potential retrieval variables were pre-selected,and ten days in-situ snow depth data of seven in-situ observations were used for analysis,Gray system theory,which has the advantage of multi-variable analysis of the small samples was selected to analyse the correlation between snow depth and retrieval parameters,and four characteristic parameters were selected based on above analysis.15 snow depth retrieval models were established.using the selected characteristic parameters,Then a comprehensive evaluation coefficient CEC of multiple regression model was defined using AIC criterion,BIC criterion and Pearson r.Then the optimal retrieval model of snow depth was selected from the above 15 models according to CEC,and the test showed the average relative error retrieval accuracy was 11.2% which was in line with operational monitoring requirements.