The erosion trend of soils is significantly associated with the physical and chemical soil properties. Wet soil aggregate stability (WAS) index is an important parameter in evaluating the erosion trends of soils. The processes of determining and preventing the soil erosion and taking measures against it are quite tiring and long-term. In recent years, estimation and modeling methods are applied to eliminate the wastages of time and effort that emerge us in the examination and investigation of soils. In this study, the soil properties which are measured routinely in semi-arid ecosystems and the predictability of wet soil aggregate stability were investigated by using artificial neural network and multiple linear regression (MLR) technique. For this purpose, 100 soil samples were obtained from 40 sample fields, and particle size distribution (sand, silt, clay), soil pH and organic matter analyses were performed. The soil properties which were selected based on the results of the statistical evaluations were used as independent variables. Accuracy of the predictions was evaluated by Mean Absolute Percentage Error (MAPE) and the root mean square error (RMSE) between the measured and predicted parameter values. The value of MAPE and RMSE derived by ANN model for WAS were 3.20 and 4.19 respectively. The corresponding values for MLR model were 13.10 and 8.32 respectively. Results showed that ANN with six neurons in hidden layer had better performance in predicting soil properties than MLR.