Cyber attacks on industrial control systems have increased dramatically due to the adoption of open, heterogeneous industrial protocols that enable remote connectivity to industrial plants. Of particular concern are attacks that impair physical processes, leading to production loss, service outages, equipment damage, human injury and environmental pollution. Given the plethora of consequences induced by cyber attacks on industrial control systems, it is vital to compute the risks associated with the attacks to enable operators to implement appropriate recovery measures. This chapter presents a probabilistic risk assessment methodology that leverages multiple Bayesian networks to measure the risks associated with cyber attacks on industrial control systems. The Bayesian networks are designed to compute the probabilities of detected cyber attacks propagating to cyber and physical assets that have not yet been compromised, thereby providing an estimate of the overall risk to the integrity of the entire industrial control system. The proposed risk assessment methodology is evaluated using a real hardware-in-the-loop case study involving a water distribution testbed.