Dr. David Stainforth of the London School of Economics took part in the Royal Society's recent Summer Science Exhibition. His exhibit was on the uncertainties involved in climate modelling. Some of his ideas are discussed in a new article on the Scientific American website.
The fundamental problem with the climate (and any long-term models of it) is this: it is chaotic. That's chaotic, in the mathematical sense. That means that the climate and certain other nonlinear dynamical systems are highly sensitive to initial conditions. Change the initial conditions, even slightly, and the long(er) terms results of your predictions diverge substantially. That's the basic reason why meteorologists are only able to offer us fairly short term predictions of the weather. The further out the prediction, the less likely it is to be accurate.
This phenomenon of sensitivity to initial conditions was discovered by Edward N. Lorenz. It's often called "the butterfly effect" because of a paper given by Lorenz in 1972 to the American Association for the Advancement of Science (AAAS) in Washington, D.C. entitled "Predictability: Does the Flap of a Butterfly’s Wings in Brazil set off a Tornado in Texas?". Lorenz's papers are a model (no pun intended) of scientific profundity and clarity (a rare combination in science). Virtually all of his key papers are available on this link at MIT's Department of Earth, Atmospheric, and Planetary Sciences. Definitely worth a look.