Changing and improving manufacturing operations in such a way that optimum flexibility is achieved is a standard task nowadays.Enhanced by the availability of CIM concepts and techniques the pervading paradigm how to solve the problem tends to be based on the structure of the data processing systems.Since management of data systems and inventory are often handled as different functional entities the complex relations of the effects of goods flows and data flows that make up the dynamic behavior of the operations as a whole often evade appropriate treatment.CIM related practice, doing the easy things first, is to follow a hands-on bottom-up approach in optimizing first individual process steps using preferably discrete simulations and then trying to add those optimized islands to a system.If we follow the original ideas of J. Forrester and his group a quite different approach is proposed. In a combination of top-down analysis and bottom-up implementation we would first apply S.D.A. with continuous simulation to understand the operations in their context as is. After optimization we would implement the upgraded system bottom-up.The approach used two levels of imaging the real system to a model. Top level simulation with a continuous model is used to analyze and define dynamic behavior, feed back loops and embedding of operations in the context of sales and supply.Bottom level simulation thereafter serves to check detailed implementation of single tasks within the dynamic specifications arrived at by the continuous overlay model.The procedure allows to exploit the strong points of both continuous and discrete simulation, namely analysis of the dynamic behavior of complex and intertwined systems of flows of goods and data on the one hand and detailed analysis of process steps involving clearly defined operations with work pieces handled.A few examples serve as illustration how this first step of a top-down optimizing with the aid of S.D.A. worked in defining manufacturing systems as a whole before starting bottom-up implementation.As the S.D.A. model is a lumped together model of the real system, its use for on-line prognosis can be a welcome byproduct.