Cause-and-effect reasoning is at the core of fault diagnosis and hazards analysis in process systems, thereby requiring the development and use of causal models for automated approaches. Furthermore, causal models are also required to explain the decisions and recommendations of artificial intelligence-based systems, lack of which is a serious drawback of purely data-driven approaches. Here, we demonstrate an approach for building multi-level causal models. A hierarchical approach is proposed to capture both cyclic and non-cyclic features of a process plant. Decoupling these features of the plant by constructing two tiers of digraphs, one tier representing overall plant and the other representing individual subsystems, helps in better inference of causal relations present in the system. An algorithm that subsides the effects of indirect causal interactions using reachability matrix and adjacency matrix ideas is also proposed. The algorithm is tested on the Tennessee Eastman benchmark process and the resulting causal model is found to represent the true causal interactions present in the system. (C) 2018 Elsevier Ltd. All rights reserved.