Soil carbon is an important factor in the process of mitigating climate change and solving greenhouse gas problems. However, the previous technology for soil carbon content analysis required a lot of labor, time, and expensive equipment (i.e. an elemental analyzer). In this study, the disadvantages of previous analysis method were secured by using smartphone images and multiple regression analysis. To predict the soil carbon content, the color variables (e.g., RGB, CIE-L * a * b * , CIE-L * c * h * , and CIE-L * u * v * ), gravimetric water content, and bulk density were used as statistical data. After Pearson's correlation analysis, several variables that had high correlations were removed and then used. In addition, the result of variance inflation factor (VIF) analysis indicated that all variables should not cause multicollinearity problems. The predictive model was classified based on land use, and the predictive model for the entire sample was also derived. The adjusted coefficient of determination (Adj. R 2 ), root mean square error (RMSE), and mean absolute percentage error (MAPE) were used to verify the predictive model. When the verification method was substituted for each predictive model, the reliability of the predictive model classified based on land use was high. Therefore, in order to predict the carbon content in the agricultural soil, it is efficient to assign each prediction model after classifying agricultural land.