AbstractMaterial Requirements Planning (MRP) systems have been widely applied in industry to better manage multiproduct, multistage production environments. Although many applications have been quite successful, much is still left to the planner's intuition as to how to assure that master schedules, component lot sizes, and priorities realistically conform to the capacity limits at individual work centers. Capacity issues may indeed be the soft spot in MRP logic.This paper explores some possible causes of irregular workload patterns when using an MRP system. Better insight on which factors cause temporary bottlenecks could help managers better assess the vulnerability of their plants to this problem. It might also suggest ways of dampening peaks and valleys. The problem setting is a multistage environment; several products are made from various subassemblies and parts. Each shop order is routed through one or more capacitated work centers. An order is delayed either by temporary capacity shortages or the unavailability of components. Of course, the second delay can be caused by capacity problems previously encountered by the shop orders of its components.Seven experimental factors are tested with a large‐scale simulator, and five performance measures are analyzed. The factors are the number of levels in the bill of material, the average load on the shop, the average lot size, the choice of priority rule, demand variability, the use of a gateway department, and the degree of equipment specialization. We have one measure of customer service, two for inventory, and two for workload. The workload measures are unconventional, since our interest is when workload variability occurs and how it affects inventory and customer service.The simulator has been developed over the course of eight years, and since this study has been further enhanced to handle many more factors. The simulator was validated recently with real data at two manufacturing plants. It is quite general, in that the bills of material, shop configuration, routings, worker efficiencies, and operating rules can be changed as desired.An initial screening experiment was performed, whereupon the average load and priority rules were not statistically significant at even the .05 level. A full factorial analysis with two replications was then conducted on the five remaining factors. Multivariate analysis of variance (MANOVA) and analysis of variance (ANOVA) statistical tests have been performed.The results confirm that workload variability can have a detrimental impact on customer service and inventory. The following structural changes to the manufacturing system can be beneficial, but tend to be more difficult to achieve. More BOM levels improve customer service, but increase inventory and capacity bottlenecks. Resource flexibility is a powerful tool to reduce workload variability. Capacity slack averaging much over 10% is wasteful, having no benefits for inventory and customer service. In general, revising the routing patterns only, such as creating more dominant paths, will not give big payoffs. The following procedural changes are easier to implement. Master schedules which smooth aggregate resources are an excellent device to reduce workload variability. Even with a smooth MPS, debilitating workload variability can still occur due to the design of the BOM, lot size, and leadtime offset parameters. Selecting a priority rule does not seem to be of overriding importance compared to master scheduling and component lot sizing. These findings must be considered within the context of the range of plant environments encompassed by this study.