With supervised learning, new environments and intrusion definitions can easily be incorporated into the system by collecting representative data and training the system on that data. The output of the training session is a set of coefficients for a new intrusion discrimination network. Once the new network replaces the previous network, the processing is ready for the new scenario. GRC recently completed final tests on this system that consisted of 44 alarm events tested by one or more humans entering the sensor field. In addition, 75 nuisance events, such as dogs, trucks, or humans traveling on the edge of the field, were incorporated. These events were recorded at the Waterways Experiment Station physical security sensor field. Except for a training set of 30 events, these conditions were not seen by GRC prior to testing. At the selected operating point, 40 of 44 alarm events, and 72 of 75 nuisance events were classified correctly. This was a significant improvement over those obtained with individual sensors with all possible logical combinations of the individual sensor outputs. 5 figures, 5 references (Author abstract modified)