It's something that comes with practice, but in general it's much better to overestimate than underestimate.

- Always take your time. Don't be rushed into plucking a number out of thin air.
- Break the task down into really small, atomic chunks.
- Each of those chunks will take at least twice as long as you think it does - nothing goes to plan 100% the first time!
- Make sure you add contingency at the end.