Modern testbeds for the experimental analysis of critical infrastructures either totally ignore the human factor or incorporate real humanmachine interfaces (HMIs) and software that require the presence of human operators during an experiment. Although experimentation with humans in the loop can provide invaluable experimental data about human decision making and reactions, it is infeasible to conduct a systematic exploration of the vast parameter space of possible human operator decisions, reasoning and actions. This paper argues that testbeds should incorporate simulated human decision making capabilities in order to engage humans in the loop, especially because humans play crucial roles in cyber security experiments involving critical infrastructures. An extension of a previously developed experimentation framework is also described; the extension provides generic "human decision" units that enable the integration of human operator and HMI models. The utility of the approach is demonstrated by assessing the impact of human operator reactions during an attack on a cyber-physical infrastructure incorporating the IEEE 30-bus power grid model.The interaction of human operators with critical infrastructures is mostly implemented using information and communications technologies, largely due to reduced costs and greater efficiency, flexibility and interoperability. Consequently, several approaches have focused on the design of (graphical) interfaces, also known as human-machine interfaces (HMIs), that assist decision making processes and reduce the reaction times of human operators [12,18]. On the other hand, human operators can also interact with a system independently of HMIs, for example, by switching a device on or off. Therefore, testbeds that focus on the analysis of critical infrastructures should also take into account the presence of human operators.This paper argues that the presence of human operators and HMIs dramatically changes system behavior and should be taken into account when designing testbeds for analyzing critical infrastructures. Existing testbeds (see, e.g., [5,11,13,21]) may engage human operators and real HMIs, but they do not include software simulations of these components. Although testbeds with human operators and real HMIs provide reliable experimental data, they are unable to support exhaustive parameter testing. This is mainly due to the costs involved in acquiring and training human operators, and the costs of customizing proprietary HMI software.Recognizing the importance of human operators and, especially, operating procedures in security experiments, we propose an extension to our previously developed experimentation framework [6]. The extension incorporates "human decision" units that can run human operator and HMI models in real time. Actions issued by the models are translated into commands that are executed in the cyber and physical realms. This way, the extended framework enables the recreation of realistic scenarios in which operators interact with the cyber realm and also ...