Watershed average slope length (ASL) is defined as the average distance measured along the main channel from the watershed outlet to the basin divide. It is an important parameter in a hydrologic model. Theoretically ASL is related to the calculation of lateral flow and routing timing, which eventually impacts simulation of the total surface runoff. SWAT (Soil and Water Assessment Tools) is a river basin scale model developed to predict the effects of management decisions on water, sediment, nutrient and pesticide yields with reasonable accuracy on lager, and ungauged river basins. The AVSWAT model, an integration of SWAT2005 and ArcView GIS 3.2, was applied in this paper to simulate monthly stream flow (1990-2009) of the Yingluoxia outlet of the 10 000 km² upstream basin of the Heihe River basin (HRB). The simulation showed that ASLs calculated by the AVSWAT model were not consistent with the actual cases in the upstream basin of HRB. As a result, it led to an inaccurate simulation. To address this problem, a new approach has been adopted to correct ASL of each sub-basin. A regression equation, which reflects the relationship between average slope and average length, was established by measuring average slope of each sub-basin. Using the corrected slope lengths and the default parameter set of SWAT together with meteorological and hydrologic data from 1990 to 2009, Nash-Sutcliffe coefficient (NSE) of monthly runoff simulation is obviously improved to 0.75 in contrast with 0.60 with uncorrected slope lengths and same parameter set. Sensitivity analysis and an examination of CN2 parameter were also conducted to prove the applicability of this correcting method. The results showed that corrected slope lengths were more suitable for the studied mountainous basin and also verified its applicability of the adopted correction approach. Using the corrected ASLs, an error-trial optimization was used to optimize other parameters. The validation with daily runoff from 2000-2009 showed an NSE of up to 0.81, which suggested an acceptable simulation for the study basin. We concluded on basis of corrected slope lengths quite good simulation can be achieved by parameter optimization. Therefore, the approach to correct ASL is feasible in mountainous basin, and the corrected slope lengths have made the stream flow simulation better.