The manufacture of complex telecommunications systems typically involves a final-assembly shop that assembles a variety of highly customized end products, each requiring many parts. Under a just-in-time (JIT) manufacturing discipline, final assembly pulls the production of parts from multiple feeder shops. The demands imposed on feeder shops are determined by the sequence of end products assembled in the final-assembly shop. To maintain a JIT discipline, such demands must be smooth over time. We describe the model and algorithm that form the basis of the final-assembly sequencer (FAS), a software tool that smooths the demands on each feeder shop. The final-assembly sequencing algorithm uses an efficient search heuristic to select a nearly optimal sequence from among many feasible alternatives. AT&T's Denver Works has implemented FAS as part of the integrated pull manufacturing (IPM) system for producing private-branch exchange (PBX) systems. Versions of FAS have also been implemented at other AT&T manufacturing locations.