Accurate detection of faults is desired to reduce risks and costs. The identification of the propagation path of faults and the system’s variables responsible of faulty operating conditions is also paramount. This paper presents a new alternative to a well-known Bayesian network-based approach to detect and identify root causes in multivariate processes. It deals with the complexity generated by the use of Bayesian network in terms of structure and decision making. The new strategy is straightforward and based on statistical foundations. The new approach revives the interest of Mason, Young and Tracy (MYT) decomposition of quadratic statistics and alleviates considerably the complexity of their upgrades. Comparison with previous approaches and performance evaluation using the Tennessee Eastman process demonstrate the feasibility and interest of the new proposal.