The improved spatial resolution of the data from the SPOT satellite provides a substantially better basis for monitoring urban land use and growth with remote sensing than Landsat data. The purpose of this study is to delineate the road network in 20-m resolution SPOT-images of urban areas automatically. The roads appear as linear features. However, most edge and line detectors are not effective in detecting roads in these images because of the low signal to noise ratio, low contrast and blur in the imagery. For the automatic recognition of roads, a new line detector based on surface modelling is developed. A line can be approximated by a piecewise straight curve composed of short linear line-elements, called linels, each characterized by a direction, a length and a position. The approach to linel detection is to fit a directional surface that models the ideal local intensity profile of a linel in the least square sense. A Gaussian surface with a direction of invariance forms an adequate basis for modelling the ideal local intensity profile of the roads. The residual of the least squares fit as well as the parameters of the fit surface characterize the linel detected. The reliable performance of this line operator makes the problems of linking linels more manageable. / Master of Science