Early detection of anomalies is crucial to maintain high productivity at a refinery. For this purpose, we propose an anomaly detection system based on adaptive resonance theory (ART) for industrial plants. The feature of the system is that it has several ART systems applied to subsystems of the plant in order to narrow down the cause of the anomalies. This report presents our examination of online anomaly detection tests on whether or not the proposed system is applicable to a distillation tower system. The tests were conducted with experimental equipment of a distillation tower in Universiti Teknologi PETRONAS (UTP). In the tests, we carried out four cases of anomaly operation (e.g., valve sticking and tray upset) that would cause quality or yield losses in the product. By learning normal operation data, the proposed system could detect anomalies in all four cases, and no false positives were observed in normal operation. We also found that the system could narrow down the cause of the anomalies by using the results of each ART system, thereby demonstrating that the system is applicable for the distillation tower system.