If you're using a chi-square or its better relative, the g-test, then outliers do not matter.
If you're using ANOVA or Student's T-test, don't do that. They assume that you're sampling from normal distributions and you are not.
What you should do is truncate extreme outliers, then use a z-test. Start at http://elem.com/~btilly/effective-ab-testing/index.html#asli... and use the arrow keys to go through about 2 sections to for details on that.
If you have any questions, shoot them to btilly@gmail.com.