In conventional approaches to scheduling problems, a single dispatching rule was applied to the all machines in a manufacturing system. However, since the condition of a machine generally differs from those of other machines in the context of overall system operation, it is reasonable to identify a suitable dispatching rule for each machine. This study proposes an adaptive procedure which produces a reliable dispatching rule for each machine. The dispatching rule consists of several criteria of which weights are adaptively determined by learning through repeated runs of simulation. A Taguchi experimental design for the simulation is used to find effective criteria weights with efficiency and robustness. For evaluation, the proposed method was applied to a scheduling problem in a semiconductor wafer fabrication system. The method resulted in reliable performances compared with those of traditional dispatching rules.