Slope spectrum is an effective method of reflecting macro terrain feature with the quantitative micro-terrain-factor, slope, which has been attracting more and more attentions in the geomorphology research area. In this paper, a method of automatic topographic recognition based on slope spectrum was proposed. First, characters of slope spectrum derived from DEM is extracted as quantitative factors to describe terrain feature. And then, this method using BP neural network model to automatically recognize landforms from DEM data quickly and accurately, according to characters of slope spectrum. In this paper, eight sample areas from different landform types of Shaanxi Province in China are selected to test this method. As is shown in experimental results, the recognition rate is 70% on average and the best combination of terrain feature factors are found out in sample areas, which shows a great potential in landform recognition.