The knowledge about spatial distribution characteristics of soil organic matter (SOM) contents is of important significance to development of the agriculture. Models different in simulation mechanism are used to predict distribution of SOM contents, because different regions have soils of different soil properties and different environmental factors affecting soil properties. It is, therefore, essential to choose an optimal suitable model site-specifically for the prediction, this paper is oriented to explore for a model that is capable of predicting spatial distribution of SOM contents in a tropical region complicated in relief, like Hainan Island. Ordinary kriging (OK), regression-kriging (RK) and random forest (RF) were the three models used to predict spatial distribution of SOM contents in the Hainan Island based on the training set of 128 samples, combined with topographic factors, normalized vegetation index, soil types and landuse types, and then the predictions were validated with a validating set of 32 validation points in the study area for comparison of prediction accuracy. Results show: (1) The optimal fitting model for prediction of distribution of organic matter contents in the 0 ~ 5,0 ~ 20, 20 ~ 40 and 40 ~ 60 cm soil layer is RF, RF,RF and OK, respectively. RK and RF can better describe local variation of SOM contents; (2) The mean value of SOM contents in the four soil layers was 19.67, 15.89, 10.30 and 8.07 g kg^-1, respectively, showing a spatial distribution trend that SOM contents are higher in the southwest and northeast parts and lower in the west and southeast parts of the island.