The scale and complexity of computer-based safety critical systems, like those used in the transport and manufacturing industries, pose significant challenges for failure analysis. Over the last decade, research has focused on automating this task. In one approach, predictive models of system failure are constructed from the topology of the system and local component failure models using a process of composition. An alternative approach employs model-checking of state automata to study the effects of failure and verify system safety properties. In this paper, we discuss these two approaches to failure analysis. We then focus on Hierarchically Performed Hazard Origin & Propagation Studies (HiP-HOPS) - one of the more advanced compositional approaches - and discuss its capabilities for automatic synthesis of fault trees, combinatorial Failure Modes and Effects Analyses, and reliability versus cost optimisation of systems via application of automatic model transformations. We summarise these contributions and demonstrate the application of HiP-HOPS on a simplified fuel oil system for a ship engine. In light of this example, we discuss strengths and limitations of the method in relation to other state-of-the-art techniques. In particular, because HiP-HOPS is deductive in nature, relating system failures back to their causes, it is less prone to combinatorial explosion and can more readily be iterated. For this reason, it enables exhaustive assessment of combinations of failures and design optimisation using computationally expensive meta-heuristics. (C) 2010 Elsevier Ltd. All rights reserved.