Large sample sizes ($n > 30$ or so): The CLT rescues you - Mild skewness: A little skewness usually doesn't cause problems - Sample means and proportions: Even if individual observations aren't normal, their averages tend to be