This paper addresses the problem of determining the batch sizes for parallel machines in semiconductor wafer fabrication which has the characteristics of re-entrant process flows. Due to complexity of re-entrant process flows, semiconductor wafer production system has high variability in arrival patterns and experiences high imbalances among processes, which eventually degrades whole system performances. In this paper, we show that finding optimal batch sizes for key processes can considerably reduce the imbalances among processes and queue lengths between processes. A simple technique for determining the optimal batch sizes for parallel machines with different speeds and variability is developed and applied to solve a case problem. The results are briefly discussed in terms of system parameters involved in manufacturing environment.