Soil temperature is one of the main characteristics of soil that its changes have a great impact on many processes such as growth, plants flourishing and soil formation. Nevertheless, temperatures throughout the soils profile are not measured continuously. As a result, we encounter the lack of statistics in soils temperature data, while meteorological parameters are being measured regularly. Since presented relationships in the previous investigations do not provide the appropriate accuracy to predict soils temperature, the objective of this paper is to introduce a high accuracy relationship based on comparison of regression methods and Artificial Neural Network (ANN) by using daily meteorological data of three stations located in Mashhad, Sabzevar and Shiraz. Solidarity coefficients indicated that ambient temperature, evapotranspiration and evaporation have the most solidarity with the soil temperature at a depth of 5 cm, respectively. According to solidarity coefficient and results of the 2 models, air temperature, evapotranspiration, humidity and effective precipitation with daytime lag of one day were regarded as the best input parameters, respectively. The results showed that second order regression with single variable had the lowest accuracy while the highest accuracy was observed in the ANN method. In the meantime, multiple regressions had a reasonable accuracy. In calculation of freezing depth we concluded that Finish equation has an acceptable accuracy, whereas by considering an added parameter related to the precipitation depth in the cold days to the equations order, the results of the Finish equations will improve dramatically. Maximum error of 15% was observed for the recommended equation in Shiraz station and count of error in Mashhad and Sabzevar 0.6% and 2% observed respectively.