The RePEc blog has a piece that should be an eye opener to everyone relying on impact factors to establish the quality of a journal. The publisher of these impact factors, Thomson Scientific, is unable to provide the data to replicate them! Using the standards of the journals Thomson Scientific is trying to measure, this would be considered fraud and obviously unpublishable.
I find this very unsettling, considering that so much in terms of tenure and promotion decisions in the academic world is relying on this. Imagine the consequences this could have in terms of litigation by those denied tenure... or by universities that were potentially misled.
I have been using these impact factors to decide where to send my manuscripts. I will not use them anymore and rather use the ones that RePEc computes. While they are clearly not comprehensive (but Thomson Scientific's are not either, and probably worse), at least they are openly computed. It looks like all the data is there on the web to replicate, there is just too much of it... but I have been able to replicate my own citation counts.