There is an excellent technique to avoid this – it is called planning poker.
Each team member gets a deck of 13 cards as shown above. Whenever a story is to be estimated, each team member selects a card that represents his time estimate (in story points) and places it face-down on the table. When all team members are done the cards on the table are revealed simultaneously. That way each team member is forced to think for himself rather than lean on somebody else’s estimate.
If there is a large discrepancy between two estimates, the team discusses the differences and tries to build a common picture of what work is involved in the story. They might do some kind of task breakdown. Afterwards, the team estimates again. This loop is repeated until the time estimates converge, i.e. all estimates are approximately the same for that story.
Note that the number sequence is non-linear. For example there is nothing between
40 and 100. Why?
This is to avoid a false sense of accuracy for large time estimates. If a story is estimated at approximately 20 story points, it is not relevant to discuss whether it should be 20 or 18 or 21. All we know is that it is a large story and that it is hard to estimate. So 20 is our ballpark guess. And no, you can’t cheat by combining a 5 and a 2 to make a 7. You have to choose either 5 or 8, there is no 7.

Home
source: http://wwwis.win.tue.nl/2R690/doc/ScrumAndXpFromTheTrenchesonline07-31.pdf