Abstract: Abstract: The soil erosion model, Universal Soil Loss Equation (USLE) model and its principal derivative, Revised Universal Soil Loss Equation (RUSLE) model have been widely used in the past decades. However, the use of USLE and RUSLE has been limited as they can't generate reliable estimation of topography factor (i.e. slope length and slope grade, LS). Due to the fact that simple slope length can not fully represent the influence of topography, various approaches and algorithms have been proposed. These algorithms are raster grid cumulation, unit stream power theory, contributing area, sediment transport index and network triangulation etc., which partly reflect the impact of complex terrain. In this paper, several approaches were discussed, which were Remortel code (its benchmark slope length being 22.13 m), Zhang Xiankui algorithms (its benchmark slope length being 20 m), Moore algorithm (representing unit stream power theory), Desmet algorithm (representing contributing area method) and B?hner algorithm (representing sediment transport index method). Based on the measured data from slope length plot and McCool's data, the influence of standard slope length was discussed, and the characteristics of different algorithms of LS factor were analyzed. It is found that when the slope gradient is relatively small, the difference of slope length (L) factor between the benchmark slope lengths of 20 and 22.13 meters is small. With the increase of slope gradient, the difference is becoming more and more obvious under the superposed impact of slope length index. In addition, L factor calculated by RUSLE algorithm is closer to that of measured data, and the comparative analysis combined with McCool data shows that the slope length's exponent is more suitable for the middle erosion ratio of rill and interrill. Given the measured data used for only one year, further varification would be needed for the fitness of RUSLE. It is also found that the referenced values of LS factor are larger than that of Zhang Xiankui algorithms, Moore algorithm, Desmet algorithm and B?hner algorithm. It indicates that from the slope of linear regression equations, the closeness between the four above-mentioned algorithms and reference values increases respectively. However, L factor and slope (S) factor don't show the same trend. Taking LS factor value from Zhang Xiankui as an example, there is larger difference for L factor between algorithms of Zhang Xiankui and McCool, while both of the S factor values are almost the same. As for the spatial distribution of LS values from different algorithms, there presented different characteristics. Raster cumulative algorithm generally exhibits more fragmentation, and there are more parallel bands on straight slope. While the other three algorithms show a smooth and continuous feature overall, but there exist areas without data.