There is a growing concern in process plant safety to assess risk. Two important motives can be identified: on the one hand, the new Seveso directive, which contains a number of risk-related requirements for process plants, and on the other, the advent of the "Safety Integrity Level" SIL classes. These call for methods of analysis, which enable one to obtain reliable results at reasonable effort. The chief obstacle against the widespread use of fault tree analysis for analyzing the safety of process plants is the dearth of appropriate reliability data. This gives rise to time-consuming and error-prone searches in the analyses performed, which nevertheless suffer from the drawback that they do not exactly reflect the situation because the reliability data is not plant-specific. This being the case, in the present approach generic ranges for reliability data are used. These are considered to reflect the uncertainty of the reliability data caused amongst others by their transfer from the conditions of origin to other plants. These uncertainties are propagated in evaluating the fault tree and yield results at both the minimal cut set and structure function levels characterized by natural language expressions. Numerical results are supplied as additional hints for system improvement. The source for the reliability data assignments is a quality assured evaluation in two plants of the explosives industry. The method is implemented in the program system Semi-Quantitative Fault Tree Analysis (SQUAFTA), which was successfully applied to the analysis of a process plant. (C) 2004 Elsevier Ltd. All rights reserved.