1) Decide beforehand how many subjects you're going to use in the study.
What is the cutoff point?
2) Have at least 20 participants per condition.
You have to have a good number of subjects in order for your results to be believable.
3) Must disclose all the variables you measured in the study.
"You can't do a study that looks good... then collect a whole lot of other stuff." In other words, don't measure a a bunch of other unrelated variables, just in case you didn't find a connection between the variables in your original hypothesis.
4) Report your results both with and without any participants you excluded.
State upfront what you would call an outlier (an extreme score) in the data.
Dr. Michael Britt does some wonderful podcasts on his website, The Psych Files. Definitely something to check out if you have the time. He also publishes apps related to psychology (I found this particular video in the "Psychology Student Survival Guide" app on the App Store).
Have a great weekend! :-)