In this research the E-C translations from a self-built corpus of College English Competence Test were first manually annotated with translation errors according to potential evaluative parameters ，then the translation errors were concordanced in the light of each potential evaluative parameter and statistically ana- lyzed for correlations with the E-C translations ’formal manual scorings respectively． Thus some statistically significant evaluative parameters that are actually functioning in E-C translation manual scorings were tentative- ly identified ，and verified by an automatically constructed evaluation model of multiple linear regression ， whose evaluation performances bear a Pearson correlation of 0 ． 78 with the formal manual scorings． So it can be concluded that the study tentatively reveals the translations ’text features that raters have been paying close attention to in current E-C translation scoring practice for college English testing from an empirical perspec- tive ，and it might as well shed some light on automatic quantitative evaluation of E-C translations in language testing．