The breakdown point of an estimator is the smallest fraction of observations one must corrupt to make the estimator take any value one wants.