Stereotomography is a robust method for building velocity models from seismic reflection data, and it has been applied to offshore seismic data, but there is almost no stereotomographic study with rugged topographic conditions. We study the topographic effects on the slopes of locally coherent events of seismic data and develop an approach to calculate the slopes on an undulant observation surface using the horizontal and vertical components of slowness vectors estimated. Then, we develop an extended stereotomography with undulant observation surface based on the conventional one. Tests on synthetic data validate the extended stereotomography. Application to the field seismic data in a foothill belt in Xinjiang of the West China indicates that the extended stereotomography is an effective tool to build velocity models for prestack depth migration of seismic data acquired on rugged topography.