Fault detection and diagnosis are crucial to reducing risks and costs in any process. The identification of the propagation path and the variables responsible for faulty operating conditions is also vital. This paper presents a causal network-based approach to detect, diagnose, and identify root causes in multivariate processes. We discuss aspects such as complexity and rules related to modeling such network approaches. The proposed strategy is established on statistical justifications. The introduced decision rules deal with unknown faults and offer new perspectives to data-driven methods for fault diagnosis. The proposed approach is evaluated and demonstrated using the well-known Tennessee Eastman Process (TEP) benchmark.