Empirical soil erosion models continue to play an important role in soil conservation planning and environmental evaluations around the world. The effect of hillslope length on soil loss, often termed the *slope length factor*, is one of the main and most variable components of any empirical model. In the most widely used model, the Universal Soil Loss Equation (USLE), normalized soil loss, *L*, is expressed as a power function of slope length, λ, as $\begin{array}{c}L=(\lambda \mathrm{/22.1}{)}^{m}\end{array}$, in which the slope exponent, *m*, is 0.2, 0.3, 0.4, and 0.5 for different, increasing slope gradients. In the Revised Universal Soil Loss Equation (RUSLE), the exponent, *m*, is defined as a continuous function of slope gradient and the expected ratio of rill to interrill erosion. When the slope gradient is 60% and the ratio of rill to interrill erosion is classified as moderate, the exponent *m* has the value of 0.71 in RUSLE, as compared with 0.5 for the USLE. The purpose of this study was to evaluate the relationship between soil loss and slope length for slopes up to 60% in steepness. Soil loss data from natural runoff plots at three locations on the Loess Plateau in China and data from a previous study were used. The results indicated that the exponent, *m*, for the relationship between soil loss and the slope length for the combined data from the three stations in the Loess Plateau was $\begin{array}{c}0.44({r}^{2}=0.95)\end{array}$ For the data as a whole, the exponent did not increase as slope steepness increased from 20 to 60%. We also found that the value of *m* was greater for intense storms than for less intense storms. These experimental data indicate that the USLE exponent, $\begin{array}{c}m=0.5\end{array}$, is more appropriate for steep slopes than is the RUSLE exponent, and that the slope length exponent varies as a function of rainfall intensity.