The distribution and variability of salinity in the world's oceans is a key parameter to understand the role of the oceans in the climate system. SSS can be retrieved from the brightness temperature of Soil Moisture and Ocean Salinity (SMOS) measurements and Auxiliary data SMOS satellite provided. This paper provides a new statistical approach for modeling SSS for SMOS data to improve the accuracy of retrieved SSS without considering/understanding the related mechanism. Based on the Bayesian Network model, for 9911 pair matchup data in training data set in May 2015 in the global ocean, the mean absolute error (MAE) of model SSS reaches 0.42 practical salinity units (psu) compared to Argo SSS, which is 1.79psu for SMOS Level 2 SSS. Validation is also made using Argo measurements in the June 2015. The MAE of model and SMOS SSS are 1.06psu and 1.55psu respectively compared to Argo daily SSS. The validation results show that the new model is effective correction to predict SSS based on SMOS satellite data.