Abstract: Oracle is one of the most difficult and expensive parts in automated software testing. It is explored in this paper to use Radial Basis Function Neural Networks (RBF NN) to construct an automated oracle model. The automated oracle generate the approximate outputs that are close to expected outputs after training. Actual outputs are then compared with the approximate outputs to determine if there is a failure when software is running. Oracle can therefore be implemented automatically and the precision be adjusted by parameters. It will save a lot of time and cost in software testing.