This paper concerns dynamic part dispatch decisions in electronic test systems with random yield. A discrete time, multiproduct, miltistage production system is used as a model for the test system with the objective to minimize the sum of inventory holding, backlogging, and overtime costs over a finite horizon. Exact results for such systems have been limited to either single-stage, multiple time period, or multistage, single time period problems with a single product. Here we develop two approximate policies: the linear decision rule, and the myopic resource allocation. The effectiveness of the two policies is evaluated through simulation under different operating conditions representative of those encountered in IBM and Tandem Computer facilities. The extensive computational study clearly demonstrates the overall superiority of the linear decision rule.