Based on a noncausal data structure, this article develops a statistical-based monitoring scheme for diagnosing abnormal situations in complex systems. The recorded variables are assumed to exhibit Gaussian and non-Gaussian signal components, which are monitored using the statistical local approach. For diagnosing abnormal conditions, the paper introduces a regression-based technique that allows estimating the fault contribution from abnormal operating conditions. Application studies involving a simulation example and the analysis of recorded data from an industrial melter process demonstrate that the proposed diagnosis scheme is more sensitive in analyzing incipient fault conditions than existing approaches discussed in the literature. (C) 2011 American Institute of Chemical Engineers AIChE J, 58: 23572372, 2012