We present a framework for developing hierarchical models for predicting system health (e.g. probability of failure within a given mission duration), based on component-level reliability and degradation models. Component models may be specified as parametric probability distributions or nonparametrically as empirical distribution functions. Flowgraph methods are then used to predict the system failure time distribution. We illustrate with an application to aircraft maintenance. Copyright © 2011 John Wiley & Sons, Ltd.