The reason why premature optimization is bad is that optimization usually leads to a worse design. Unlike refactoring, which leads to a better and cleaner design, if done thoughtful and right. What I learned to be useful for me to analyze the usefulness of a refactoring was first looking at our UML diagram to visualize the change and then writing the code-doc (e.g Javadoc) for the class first and adding stubs ahead of any real code. Of course experience help a lot with that, if in doubt, ask your favorite architect ;)
