Models can be validly defined at many levels, from general concepts to specific mechanisms. Consider for example Marr's and Biederman's models of vision by abstraction of features; Selfridge's Pandemonium model; Triesman & Gelade's spotlight theory of attention; Collins & Quinlan's Spreading Activation model, McClelland & Rummelhart's PDP approach, to name just a few. Some of these models are far more general and conceptual than mine, but are valid models nonetheless. And then of course there is the example of the Gestalt view of perception, a concept so general that even I would hesitate to call it a model. Despite its vagueness as to specific mechanism however, the Gestalt view of perception served as an invaluable reminder to focus on the more difficult aspects of perception, and to resist the temptation to consider only simpler aspects of perception that can be described by specific models. It would have been a great loss if the Gestalt ideas had been denied publication because the concept was not sufficiently specified! General models are appropriate in a new field where much remains to be discovered, while more specific models are derrived as more exact specification of general models, as a science matures. Whenever a fundamentally new direction is proposed, it is often presented initially as a general model which is only refined later as the details of the approach are worked out. It has been my intention to take the very general and vague Gestalt notions such as the soap bubble analogy, and to advance them one step closer to a complete computational model. As long as my model is presented in more concrete and specific terms than the original Gestalt principles themselves, it must be seen as an advance on those principles. If every detail of the mechanism has not been explicitly defined, that does not invalidate the general concept, and conversely, presenting a partial or incomplete specification of the computational mechanism in no way invalidates the general concept. While I do not present detailed specifics of a model, I do propose a specific approach, with specific and clearly stated principles dictating how such a model should be defined in order to incorporate those computational principles, and I describe an outline of a specific mechanism to help evaluate whether the concept is viable in a physical system.