| This study is intended to contribute some empirical evidence to the authenticity on proofreading of NMET by the corpus-based approach. As an important factor that affects validity, authenticity is investigated from two perspectives in this study: one is how authentic the proofreading items in NMET are; the other is how the degree of authenticity affects students' test performance.The errors designed in the NMET proofreading items were classified according to CLEC (Chinese Learners English Corpus), and the error types were compared with chose in CLEC. The general statistical results showed significant correlation between NMET and CLEC, which indicated that the error types in the NMET proofreading items hoiistically represented the ones students often made in their writing.To investigate the effect of authenticity on students' test performance, authenticity was categorized into four levels according to different levels of representativeness in language points and error types by corpus-based approach. The language points, as represented by the correct answers in the proofreading items, were concordanced in BNC and partitioned into two kinds by their frequencies: high and low levels. The errors were also classified into two kinds according to CLEC: high and low one. After an experiment of proofreading test was carried out, ANOVA analysis showed that these four levels of representativeness widely varied significantly between any two kinds. However, when the language points were in low level of authenticity, the difference was not significant in various authentic degrees in error types on students' test performance. Besides, students obtained the best performance when language points were more authentic but error types were less.Through these results of comparison, some implications could be offered for test developers during test designing from the consideration of increasing more errortypes in test development, and the perspective of combination of authentic levels with testing difficulty. |