Traditionally, discrimination has been understood as an active process, and a technology of its procedures has been developed and practiced extensively. Generalization, by contrast, has been considered the natural result of failing to practice a discrimination technology adequately, and thus has remained a passive concept almost devoid of a technology. But Traditionally, many theorists have considered generalization to be a passive phenomenon. Generalization was not seen as an operant response that could be programmed, but as a description of a "natural" outcome of any behavior-change process. That is, a teaching operation repeated over time and trials inevitably involves varying samples of stimuli, rather than the same set every time; in the same way, it inevitably evokes and reinforces varying samples of behavior, rather than the same set every time. As a consequence, it is predictable that newly taught responses would be controlled not only by the stimuli of the teaching program, but by others somewhat resembling those stimuli (Skinner, 1953, p. 107ff.). Similarly, responses resembling those established directly, yet not themselves actually touched by the teaching procedures, would appear as a result of the teaching (Keller and 'Preparation of this paper was supported in part by