This paper builds on the lessons from recent financial crises to develop an analytically tractable risk management metric that more accurately measures potential exposures to financial crises and also captures volatility during non-crisis times. We develop a multiple-regime stress-loss risk framework that assumes markets are characterized by quiescent (non-crisis) periods most of the time; interspersed with infrequent crisis periods where 4-5 sigma events can occur with non-negligible probabilities. The framework is flexible and can incorporate an arbitrary number of crises. One of the primary lessons of 1998 and 2007 is that returns can be correlated due to the capital underlying a collection of trades (or strategies), regardless of any underlying economic rationale. This is an important feature of our model. We include crises that are directional in nature and capture severe directional moves such as those which occurred in 1994 and 1987, and we incorporate crises that capture strategy-based (or trade-based) crises such as occurred in 1998 and 2007. We show how the model can be used to decompose the risk of a portfolio between crisis and non-crisis risk, and how to decompose the strategy (or individual asset) contributions to the two types of risk. The model is also used, in a Black Litterman spirit, to examine the expected returns that are consistent with a given portfolio allocation and how expected returns would have to change to justify a portfolio tilt away from an initial allocation. The paper discusses the practical implementation of the model in the context of a fund of hedge funds manager.