Gradient Slope, as a basic topographic parameter, is widely used in many research fields such as agrology, meteorology, ecology and hydrology, etc. However topographic parameters extracted from DEMs have distinct scale effect, which can lead large degree uncertainty to application results. This paper investigates the scale effect of slope derived from DEMs and builds its transforming model. The research takes 48 sample sites as test areas which are located in different landforms in loess plateau of North Shaanxi province, and a series of multi-scales DEMs as test data, slope scale effect is investigated in both point slope and mean slope respectively. The experiment shows that each point slope fluctuates and decreases with the decreasing of DEM resolution. It presents the characters of both local fluctuation and global decreasing. While mean slope varies inversely and regularly with the increasing of DEM grid size. In certain range of DEM scales, mean slope meet the law of fractal geometry. Fractal dimension has relationship with the elevation differences and complexity of landforms. According to the slope scale effect, slope scaling model is established, which can effectively realize slope downscaling. The scaling result shows that those downscaled slope can efficiently approach target value on the aspect of spatial distribution structure, statistic index and slope histogram. More work should be paid mainly on solving transformation errors in the area of mixed landforms and the errors of slope histogram with two peaks.