It is important to derive an efficient query processing strategy in distributed database systems since their performances critically depend on the strategies. Join is a frequently used operation and the costs to process it are high. Thus, Join query optimization has received great attentions during past decades. Though the fragmented distributed database systems on local area networks have many advantages and are widely used, not many join optimization algorithms deal with them. In this thesis, we propose an efficient join optimization algorithm for the joins between two fragmented realtions in a distributed database system on a nonbroadcast fast local network. Our objective is to minimize the response time of a join. Semantic information associated with fragments is used to eliminate unnecessary processings. Duplicate copies of each fragment are considered to increase parallelism and to reduce the costs to process a join. A join is transformed into a union of the fragment joins and then each fragment join is assigned to a site. The similarities between the fragment join assignment problem and the balanced graph partitioning problem are