As process industries grow in complexity, detection, diagnosis and correction become more and more difficult owing to the large number of variables that have to be monitored. Many trouble shooting problems can be effectively handled by monitoring only a small subset of the entire set of process variables. In this paper we propose a metric based on Bhattacharyya distance to measure the extent of similarity of a particular fault operation w.r.t the normal in any subset of variables. Then using this similarity metric, a scheme is proposed to systematically identify a small number of key variables for a particular fault. The effectiveness of the proposed approach is demonstrated through the benchmark Tennessee Eastman Challenge problem and is compared with other distribution and distance based metrics in literature and PCA based contribution charts.