In this paper, we present a new method for robot programming: view-based teaching/playback. It was developed to achieve more robustness to changes of task conditions than conventional teaching/playback without losing its general versatility. Our method is composed of two parts: teaching phase and playback phase. In the teaching phase, a human operator moves a manipulator to achieve a manipulation task. All the movements of the manipulator are recorded. All the images of the teaching scenes are also recorded by a camera. Then, a mapping from the recorded images to the corresponding movements is obtained as an artificial neural network. In the playback phase, the motion of the manipulator is determined by the output of the neural network using camera images. We applied this view-based teaching/playback to pushing tasks by an industrial manipulator. After multiple teaching demonstrations, pushing an object to the goal position was successfully achieved from some initial positions that were not identical to those in the demonstrations, without object models or camera calibrations.