In order to make sure there's really something there. In theory, a single transit could be "faked" by having some kind of object pass between Kepler and the target star. By observing three full transits, it can be determined whether or not they are the real thing or just false alarms. As a prevoius poster mentioned, if you compare the orbital timing, and find three transits that have a nearly identical period, then it's safe to assume it's not a false alarm.

Funny, NASA seems to think that it needs 4 transits. Here is an excerpt from one of their webpage;

Considering that we want to find planets in the habitable zone, the time between transits is about one year. To reliably detect a sequence one needs four transits. Hence, the mission duration needs to be at least three and one half years.

http://kepler.nasa.gov/about/

Oh well, I like your number better.

---

Life is short so, live it well.

Three transits...four transits...same principle applies. :-P

---

You will be assimilated...bunghole!

<snip>

Considering that we want to find planets in the habitable zone, the time between transits is about one year. To reliably detect a sequence one needs four transits. Hence, the mission duration needs to be at least three and one half years.

I think the key word is reliably for 4 transits.

Of course the more transits the more accurate the results become.