We describe the architecture of an anomaly detection system based on the Hidden Markov Model (HMM) for intrusion detection in Industrial Control Systems (ICS) and especially in SCADA systems interconnected using TCP/IP. The proposed system exploits the unique characteristics of ICS networks and protocols to efficiently detect multiple attack vectors. We evaluate the proposed system in terms of detection accuracy using as reference datasets made available by other researchers. These datasets refer to real industrial networks and contain a variety of identified attack vectors. We benchmark our findings against a large set of machine learning algorithms and demonstrate that our proposal exhibits superior performance characteristics.