The standard errors of some paramters in global curve fitting seem too low (confidence intervals too narrow).

There is a bug in Prism 4.00 and 4.0a. When you choose global curve fitting (sharing the values of some parameters between datasets), Prism incorrectly reported the SE and CI of the nonshared parameters (using the sum-of-squares for that one data set but the df for all data sets). Fixed in 4.02 (Windows) and 4.0b (Mac).