SUMMARYThis paper describes procedures to build custom-tailored behavioural models of cellular neural networks (CNNs), and acompanion tool to run these models. The main property of the CNNs is the emerging behaviour, i.e. new phenomena arise from the interactions of thousands of identical cells. The existence of these phenomena need is to be checked during the design phase, which requires a full network simulation and therefore constitutes a very time-consuming step of circuit veriÿcation. To solve this task as a modelling problem, we introduce a new behavioural model optimization technique. Starting from a user-deÿned set of block models, the proposed framework produces an optimized selection which is used to build up a full-chip model. The optimization goal is the minimization of the simulation CPU time and the maximization of the time domain precision. A dedicated environment has been developed for e cient numerical simulation; this environment is brie y described in the paper. Two case studies are also presented to demonstrate the e ectivity of the technique.