This paper gives an example of a holistic, quantitative dependability analysis of a software-intensive system. It demonstrates a way of choosing costeffective improvements and achieving trustworthy results, even when it may be difficult to get agreement on software reliability data. This is done by identifying those components that are critical for the safety of the whole system. In the case analysed here the software, though extensively used in normal operation, is not critical for the accident scenario considered. Therefore a rough estimate of its failure rate suffices for dependability calculations.