Abstract In intermittent manufacturing of complex, assembled products, subassemblies are commonly produced on separate production orders, stored, and withdrawn as needed to make higher level assemblies. An alternate procedure is to make subassemblies as a set of operations on the production orders for their parent assemblies. If the latter method is employed, we say the subassemblies are combined. Whether subassemblies are made to stock or combined affects the number of production orders issued, the amount of stockkeeping and materials handling, the number of setups, and the size of the inventory investment. As a result, the impact on manufacturing costs can be substantial. In earlier work, the author developed and applied cost minimization methods for making these decisions in a preMRP, order point-order quantity environment. In this paper these methods are extended so that they are applicable to computer-based MRP systems. The proposed decision-making strategy is to select a product line and establish the set of subassemblies within the line which could be either made to stock or combined. A typical master production schedule is selected and requirements are determined for the level 1 assemblies. Each level 1 assembly is analyzed to determine the cost associated with combining it or not combining it with each of its parents. The minimum cost alternative is selected and the process is repeated for the level 2 assemblies and so forth until all levels have been analyzed. A number of numerical examples are presented including cases where subassemblies are common to multiple parents, multiple units of a subassembly are used in one parent, and subassemblies have repair demand. Also a method is developed and illustrated for making decisions where learning affects the cost of assembly operations and must be taken into account. Analysis of product structure to determine whether subassemblies should be made to stock or combined offers the possibility of substantial cost savings. This paper presents a simple, straight forward method for making the analysis in an MRP environment.