1. Read the literature extensively.
2. Identify a key "given" in the field that has never been accurately tested (yes, there are LOTS of them apples).
3. Design a PERFECT experiment to test the given dogma (PS from experience: Its almost certain that you will either find the dogma entirely wrong, or find major loopholes in it).
4. Believe in the data.
Unfortunately, implementing this four-point rule is harder than it seems (I'm still trying). Good luck!