Cellular Manufacturing Systems Design is a complex problem because it deals with multiple criteria in an uncertain environment where there are many rapid fluctuations. The generation of a consensus CMS structure, which can satisfy different criteria, may respond better the unpredictable perturbations. We have considered this first problem as searching the consensus structure in the space of consistent structures. The second problem is related with the robust elements, which form the nucleus of the system. For that a five knowledge-level approach considering the expert knowledge is presented. In the first level, the measurable criteria are selected from the set of manager criteria. In the second level, for each criteria, some consistent structures are generated by a generator. Based on the knowledge representation from criteria to machines and parts in the third level, the decision models for consensus and robustness searching are developed in the forth level. In the last level, discriminant analysis is used to evaluate each machine and/or parts from consensus and robustness points of view. An application has shown the interest of this approach.