Natural setting usability study

What do you believe is the most effective way to avoid usability study bias when conducting a controlled or natural setting usability study?