The need to analyze high-impact low-probability events on power systems, due to natural and man-related threats calls for comprehensive approaches to security assessment, which exploit the concept of risk. The paper describes a probabilistic approach and a tool, developed in the EU research project AFTER, aimed at selecting the most critical contingencies to be analyzed in depth, starting from short-term probabilistic models of incumbent natural or man-related threats and from component vulnerability curves in an integrated power and ICT system. A fast estimate of the contingency impact is obtained by using topological metrics. Results on a realistic power system include the sensitivity analyses of different contingency selection options, and the verification of the performances of the estimated impact metrics for screening purposes.