We developed velocity models to prestack depth migrate two seismic lines acquired in an area of complex mountainous geology in southern Alberta, Canada. Initial processing in the time domain was designed to attenuate noise and enhance the signal in the data. The prestack and poststack time-migrated sections were poorly focused, implying the velocity models would be inadequate for prestack depth migration. The velocity models for prestack depth migration, developed by flattening reflections on common image gathers, ineffectively imaged the complex geology. We developed our most effective velocity models by integrating the mapped surface geology and dips, well formation tops, geological cross sections, and seismic-velocity information into the interpretation of polygonal areas of constant velocity on several iterations of prestack depth-migrated seismic sections. The resulting depth-processed sections show a more geologically realistic geometry for the reflectors at depth and achieve better focusing than either the time-migrated sections or the depth sections migrated with velocity models derived by flattening reflections on offset gathers.