We use co-word analysis in a retrospective study of the transformation of the knowledge network in the field of polymer science from 1973 to 1976. The results of this study lead us to propose a model of change in the field. This model is based on the observation that the interaction of several networks gives rise to a sub-network that is at first central and then - and this is what the model allows us to predict — central and developed (without its precise content being predictable). Such sub-networks begin in regions of the network of central associated words where there are numerous holes or incomplete links. The model appears to be sufficiently robust statistically that it does not miss significant transformations and it suggests a way of predicting knowledge development. A comparison is made with other models of network transformation, such as the contagion model and the model of local structural equivalence.