Abstract. This paper analyses the robustness of self-organizing (engineered) systems to perturbations (faults or environmental changes). It considers that a self-organizing system is embedded into an environment, the main active building blocks are agents, one or more self-organizing mechanisms regulate the interaction among agents, and agents manipulate artifacts, i.e. passive entities maintained by the environment. Perturbations then need to be identified at the level of these four design elements. This paper discusses the boundaries of normal and abnormal behaviour in self-organizing systems and provides guidelines for designers to determine which perturbation in which part of the system leads to a failure.