The study shows that translation studies (TS) as undertaken by various linguistic schools from the 1950s to the early 1990s barely had anything to do with contrastive analysis (CA), which implies that Munday’s claim that already in the 1960s there was a strong link between CA and TS is not correct. It was only in the late 1990s that CA met TS and a strong link got established between the two disciplines bridged by corpus linguistics, but the real “strong link” remained undiscovered. Though CA met TS at the right time in the late 1990s and converged to the present trend of corpus-driven CA-approach to TS, the approach has been trapped in three critical problems:mishandling of the corpora, undercutting of the internal logic between CA and TS, and potential distorted results caused by translational data. To overcome these difficulties, this paper proposes an alternative approach calledcorpus-tested CA-approach to translation studies (C-CATS), which is featured by the typical empirical cycle of observation, induction, deduction, testing and evaluations. C-CATS requires both comparable corpora and translation corpora to be involved separately for CA and TS moments, ensures the real “strong link” between CA and TS which can be reduced to the entailment law as “if p, then q”, and guarantees the proposition to hold with the corpus-tested methodology.