Adapting the fractal analysis method and raster DEMs of 25 m×25 m, this paper probes into the spatial distribution of landform patterns in the loess landforms of Northern Shaanxi Province by calculating the fractal dimension and stability coefficient of stream-network. Experimental results show that: (1) The complexity of fractal structure differs with various types of loess landforms, embodied mainly in diverse fractal dimensions and stability coefficients. Among these, loess hill and ridge regions possess the highest fractal dimension and the lowest stability coefficient, reflecting its most complex geomorphological structure in the area; loess tableland and scattered loess tableland regions rank the second; and the transitional regions between loess and desert, with rather smooth and simple terrain, present the lowest fractal dimension and the highest stability coefficient. (2) It is proved that the fractal parameter based investigation on the geomorphic features of loess landforms is feasible, and DEM datasets can play a critical and fundamental role in the process. The outcomes derived from fractal dimension show a strong correlation to geomorphic type or geologic structure to some extent. Moreover, more attentions should be paid on its geographical mechanisms.