This paper presents a novel stereo matching algorithm using wavelets to obtain a dense disparity model for DEM generation from complex SPOT stereo image pairs of a region in Hong Kong. The SPOT stereo images are first adjusted to derive the corresponding epipolar image pairs such that the matching points lie on the corresponding epipolar lines of the two images. A new edge detector based on bi-orthogonal wavelets is then used to extract the minima images from the epipolar image pairs. The edge features and the dense disparity estimate are derived from the minima images using a novel hybrid stereo matching (HSM) method that combines both feature-based and area-based approaches. Incorporating the dense disparity estimate, HSM then generates a final dense disparity map using the epipolar image pairs. The results obtained indicate that this algorithm can improve 3-D topographic accuracy better than four pixels with a fewer number of control points needed.