Estimation is inherently problematic - Pitched

There is a pretty good article which describes some of the inherent problems with estimates, and that a more statistical approach is needed. It still recognises that some form of estimate is needed and advocates using the t-shirt size approach to estimate features on a broad basis. It still misses one important point in that estimates should not just concentrate on duration! What about the complexity? Just saying something is a Medium (translates as 90% of time it will be done within 3 days!) excludes the fact that it might only be able to be done by the office genius!

Another thing that is missing from most people's estimates is a level of confidence. The accuracy of an estimate depends on when you ask. Ask a developer who has spent 10years working in the domain how long it will take to tweak an existing feature and you'll have a high level of confidence... ask him about an industry he doesn't know or to use an unfamiliar technology and you'll struggle to get anything except an “I dunno”.

This is where the “cone of uncertainty” can be used as a useful visual tool. I would suggest having it printed large and stuck on the wall. Every time you are asked for a time estimate you should point at the cone and say where you think you are on it. It is the developer's best defence to the statement “but you said it would take a week!”. After a while management will start to use estimates for what they are and not as deadlines.