Currently, the problem of anomaly detection in cyber-physical systems is highly researched, and various deep learning-based techniques have been proposed to solve this task. However, apart from efficient anomaly detection, it is often required to explain the anomaly’s cause. As it is a common approach in anomaly detection to represent readings of each sensor or actuator as a separate attribute of the analyzed feature vector, the output of the model explanation techniques could be used to identify a subset of sensors and/or actuators that demonstrate anomalous activity at some point of time. In this research, the authors investigate the most commonly used model-agnostic explanation methods, namely Local Interpretable Model-agnostic Explanations and SHapley Additive exPlanations. The experiments were performed using the Secure Water Treatment dataset that models the functioning of the water treatment plant. The one-class deep autoencoder was used as an anomaly detection model. The obtained results allow concluding that the SHapley Additive exPlanations method outperforms the Local Interpretable Model-agnostic Explanations method, and it demonstrates a low level of accuracy in the generated explanations.