In this paper, a new layered cellular manufacturing system is proposed to form dedicated, shared and remainder cells to deal with the probabilistic demand, and later its performance is compared with the classical cellular manufacturing system. In the layered cellular design, each family may need more than one cell to cover capacity requirements. The proposed approach for layered cellular design involves five stages: (1) product clustering, (2) identifying number of cells and demand coverage probabilities, (3) determining cell types using the proposed heuristic procedure, (4) performing simulation to determine operating conditions and (5) statistical analysis to pick the best design configuration among layered cellular designs. Simulation and statistical analysis are performed to help identify the best design within and among both layered cellular design and classical cellular design. It was observed that as the number of part families increased, the number of machines needed to process the parts decreased first. Then the number of machines started to increase once again as the number of part families continued to increase. Another observation was that the average flow time and total WIP were not always the lowest when additional machines were used by the system. The last and the most important observation was that the layered cellular system provided much better results than the classical cellular system when high demand fluctuation was observed.