Fractal characterization is an effective way to investigate the soil surface roughness. An appropriate fractal dimension computation method is critical to obtain the accurate results. Four methods commonly used for computing the fractal dimensions were evaluated by using the W—M fractal profile curves. The results showed that the mean-root-square method had the highest accuracy in describing the soil surface roughness with fractal dimension between 1.2 and 1.6, it had a good adaptability in characterizing the real measured soil surface roughness. Based on the fact that the mean-root-square difference of surface roughness was proportional to the section scale, the mean-root-square method was improved. Compared with the theoretical curves, the new method showed a higher accuracy and provided a new method for describing the complex characteristics of soil roughness.