Estimating development time of new software has always been an inaccurate science (did I say science!) for two main reasons. Firstly it isn't always obvious what is actually required until much further down the line even when in depth analysis and interviewing has been carried out. Secondly we developers are often using tools that do not perform as advertised on the box! and just one bug could easily multiply development time

However this is where experience has its way, and we get it right more often than not. Cost becomes an issue at this point. If a one off program is to be developed then one day is often enough to determine and specify a requirements specification. For a project, a full analysis will be required and could take up to half the total development time. Then there is testing..... a very necessary albeit tedious exercise