The probability of undetected error after using a linear code to correct \nerrors is investigated. Sufficient conditions \nfor a code to be t-good or t-proper for error correction are derived. Applications to various \nclasses of codes are discussed.