Replaces the values with their ranks before measuring correlation. It picks up any tendency for one variable to rise with the other even when the relationship is not a straight line, and it is far less disturbed by outliers.