Dispatching rules are a popular and commonly researched technique for scheduling tasks in job shops. Much of the past research has looked at the performance of various dispatching rules when a single rule is applied in common to all machines. However, better schedules can frequently be obtained if the machines are allowed to use different rules from one another. This research investigates an intelligent system that selects dispatching rules to use on each machine in the shop, based on a statistical description of the routings, processing times and mix of the jobs to be processed. Randomly generated problems are scheduled using permutations of three different dispatching rules on five machines. A neural network is then trained by using a commercial package to associate the statistical description of each problem with its best solution. Once trained, a network is able to recommend for new problems a dispatching rule to use on each machine. Two networks were trained separately for minimizing makespan and the total flowtime in the job shop. Test results showed that the combination of dispatching rules suggested by the trained networks produced better results for both objectives than the alternative of using the one identical rule on all machines.