An observation that, if removed from the data, substantially affects parameter estimates (e.g., of the mean).