In Digital Surface Model generation applications, obtaining stereo correspondences is a crucial step. The geometry of stereo image generation causes significant disparities between match points. Unlike the pinhole camera model, in pushbroom sensor images, search is less structured, since there exists no epipolar geometry. This situation brings significant computational load over the stereo correspondence generation step. In this study, a fast sparse stereo pair generation scheme is presented. This approach uses SRTM data, RPC coefficients and quadratic polynomial interpolation to generate an estimate for the geodetic coordinates of each pixel in the reference image. Harris corner detector is used to generate many feature points in the reference image. The geodetic coordinate estimates at these feature points are projected onto the second image using RPC coefficients and this projection is taken as an initial estimate for a pyramidal Kanade-Lucas-Tomasi (KLT) optical flow estimation method. Since the initial estimates for optical flow are quite reliable, the resulting match points are obtained using more precise spatio-temporal derivatives during optical flow estimation step. The pyramidal nature of the optical flow estimation method allows quite fast and promising results, even for cluttered and perspective-distorted regions, which are usually difficult to mark manually. On satellite images, the accuracy of the geodetic coordinate estimates and the success of the proposed pair generation scheme are presented through experiments. The differences for urban and rural areas are also investigated. The limits of the KLT algorithm for satellite stereo match generation and the potential of the method are discussed. Based on the experiments, it is shown that the proposed approach gives quite reliable sparse stereo correspondences for urban and rural areas in relatively small execution time.