You make observations, write theories to fit them, try experiments to disprove the theories and, if you can't, you've got something.