Fixed bug: Using AIC to compare different non-linear regresson models, Prism 5 seems to suggest that the model with the lower probability of being correct is the the preferred model to use.

This was caused by a bug in Prism 5.00 (Windows) -- it has been fixed in 5.01 (Windows, update here) and 5.0b (Mac). In choosing most likely model using AIC. Prism reports the likelihood that each model is correct, and 5.00 reported these values correctly. Prism also states which one is most likely to be correct (and graphs that one). In rare situations, 5.00 chose the less likely model. This is obvious when you look at the two probabilities. 

In other words, Prism 5.00 always reported the probabilities correctly, but sometimes got the conclusion (which model to prefer) backwards.