In this paper, we study a new method for robot programming: view-based teaching/playback. It can achieve more robustness against changes of task conditions, such as initial positions of the object, than conventional teaching/playback. The proposed method is composed of two parts: teaching phase and playback phase. We used range images instead of conventional grayscale images, which were used in our previous studies. In application to pushing tasks by an industrial manipulator, view-based teaching/playback with grayscale images failed to carry an object to a goal in the case of changing lighting conditions or color of the object. On the other hand, view-based teaching/playback with range images succeeded even in such changes. The method with range images was also successful in pushing from initial positions that were not identical to those in the teaching phase.