A central theme to the preceding discussion of natural climatic variability is that the period of instrumental observations is too short a realization on which to base predictions of future climatic change. The paleoclimatic record demonstrates that many aspects of natural variability important to society are not known well enough to anticipate possible deleterious surprises. As discussed above, recent paleoclimate discoveries have focused the attention of the atmospheric sciences community on the possibility that the climate system operates in at least two quasi-stable modes, and can switch from one mode to another in decades or years [ Broecker, 1987; Manabe and Stouffer, 1994]. A comparison of possible future climate change senarios [ Houghton et al., 1992] with the paleoclimatic record also indicates that human-induced greenhouse warming may occur at a rate that is unprecedented, and drive the climate system to a state unlike any of the past [ Overpeck et al., 1991]. This possibility highlights the need to study climatic surprises of the past as the only empirical guide to how the climate system responds to large changes in climate forcing [ Broecker, 1987]. Predictive models must be able to simulate the range of past climate system responses before we can trust then to anticipate future surprises the climate system may have in store for us.