In this paper an algorithmic approach which is composed of digraphs, taboo search and simulation is proposed for loading cellularly divided manufacturing systems (CM). Generally, manufacturing cells are designed to process a family of parts which are similar to each other based on some design and/or processing requirements. Under this situation, every part has a known target cell, so there is no specific decision making for assigning parts to a cell. But, after a period of time the initial part families may not be valid any longer, because of the design and demand changes. Under this situation the problem of cell loading (or part assignment) occurs which is the determination of, to which cell (or cells) a part should be assigned to meet required system's performance levels. Manufacturing cells are usually considered as independent units inside a factory and interaction between them is tried to be minimised. Because, increasing interaction between cells means increasing complexity of the system and loosing many advantages of cellular manufacturing systems. So the cell loading problem should be considered from this perspective and the loading alternatives which results the minimum interaction between cells should be determined and the one which closely meets performance requirements should be found. Such an approach can increase the efficiency of CM systems under dynamic manufacturing situations. In the paper, the developed methodology for this problem is explained with an example application.