Realizing the difficulties involved in direct measurement of soil properties, in recent years, alternative methods have been employed. In the present research, soil texture, organic carbon, saturation percentage and lime as readily measurable parameters, wilting point, field capacity, cation exchange capacity as well as bulk density, as predicted variables were evaluated. The data set was then divided into two subsets for calibration (80%) and testing (20%) of the models. For a prediction of the mentioned parameters, neuro-fuzzy, artificial neural network and multivariate regression were applied. In order to assess the models, some such evaluation parameters as root mean square, average error, average standard error and coefficient of determination were taken into account. Results revealed that the neuro-fuzzy model gives a more appropriate estimation than the other techniques for all the characteristics where this model increased accuracy of predictions for about 34, 10, 78 and 5% for FC, PWP, CEC and BD respectability. Next after neuro-fuzzy model, artificial neural network was of a higher accuracy than multivariate regression.