Most work in urban modelling to date has proceeded under the tacit assumption that there are mechanisms of urban process, either deterministic or nondeterministic, which can be approximately described in models. An alternative hypothesis is suggested proposing that urban process as such cannot be objectively specified, although urban system behaviour may appear deterministic or nondeterministic according to the frame of representation chosen by the modeller. Besides explaining the paradox of the relative predictive success of certain simple deterministic evolution, this hypothesis also throws light on the reason why physical analogies may work in the field of urban studies. This latter point is elaborated by means of an illustration involving the 'Brussels paradigm' of bifurcation. It is suggested that predictive ability and success of analogical transfer both depend on the presence in a model of a `logical prior', a formal element which, being nonempirical, maintains its validity, not only in repeated applications across time and space, but also in transfers from one substantive field to another.
