Creativity requires the generation of novel ideas, a process often called insight. A model of this process is proposed. At the macro-level, the model is a standard symbolic architecture for problem solving. Its problem perception components consists of a layered network of processing units. At the micro-level, each such unit propagates its computational result along a subset of its forward links. The subset is selected by matching the activation levels of the links to a threshold for forward propagation. The activation levels are subject to feedback adjustment and re-distribution. A simple simulation demonstrates that these characteristics suffice to produce sudden alternatioins in a unit's output. The question whether a network consisting of such units will exhibit this behaviorr is raised.