A new approach to classifying geomorphological slope units from digital elevation and remote sensing data is developed and implemented for a moderate to high-relief environment in southwest Yukon, Canada.  Geomorphological slope units are larger than the arbitrarily sized pixels in the digital database.  The main hypothesis is that higher classification accuracy can be achieved by using variables that describe complete, aggregated slope units instead of variables that describe only single pixels.  Based on the catena principle, slope units on downslope profiles are separated by breaks of slope.  Using a digital elevation model (DEM) generated from stereoscopic SPOT satellite images, slope units are identified and extracted after locating breaks of slope.  Each aggregated slope unit, consisting of a number of individual pixels, is summarized by suites of variables describing shape, topography, spectral response, topographic variability, and spectral variability.  Discriminant analysis shows that aggregated slope unit variables are effective for distinguishing slope unit classes, with over 90% of evaluation objects correctly classified.  An increase of 15-20% in maximum accuracy is achieved with variables describing complete slope units in comparison to the variables describing single pixels.  The study is a test of a DEM derived from stereoscopic SPOT images and three derivative topographic surfaces.  Field measurements are used to evaluate slope gradient, incidence value, and profile curvature.  Window-based filters are tested to reduce the mean error of the derivative topographic values, and it is shown that the DEM can be used for the required geomorphometric analysis.