A novel predictive risk index (PRI) is proposed as an indication of safety performance in the process plant. The predictor is developed based on a regular observation of unsafe acts and conditions. The unsafe observation results are quantified by a simple rating based on estimates of probability of danger (PD), frequency of work exposure (FE), number of persons at risk (RN) and maximum of probable loss (MPL). The rating are combined according to the following geometric average to give the risk index RI = (4)rootPDxFExRNxMPL The PRI is then defined as the moving average of 15 consecutive observations. A database of unsafe observations from a petrochemical plant was analyzed and compared to recorded incidents. For PRI value stay persistently below three, the plant is in safely managed condition and incident rarely occurs. For PRI value greater than three, the plant is in warning condition and incident may occur. The PRI trend successfully predicts the occurrence of several incidents in the plant. Three modes of incident occurrence were also identified. It is believed that the present PRI will be a useful tool for the safety management in the process plants. (C) 2004 Elsevier Ltd. All rights reserved.