[mmcdermott]

I loved McConnell's books, having read Code Complete, Software Estimation and Rapid Delivery.

Besides what is covered in those books, I've found it extremely useful to document assumptions. Every single estimate has some mental model of the project to be done. Code to be reused, vendors to integrate and, most importantly, things that won't be done. The real project almost always breaks with some of those high-level assumptions, but that tends to be lost in the shuffle.

Attaching assumptions to the estimate makes it much easier to do a post-mortem.

A powerful memory cannot compare with pale ink.