Previous research has shown that the direction of the cognate facilitation effect (CFE) can disappear if identical cognate words are removed from the stimulus list while keeping task requirements constant (Comesaña, Ferré, Romero, Guasch, Soares & García-Chico, 2015). These results do not fit well with leading computational models of bilingual word recognition (BIA+, Multilink), according to which there are no top-down influences at early stages of word processing. Influences would be post-lexical in nature and would result from competition at the response level. This study aimed to examine this issue by manipulating stimulus list composition and examining its impact on cognate word recognition. We varied the proportion of identical cognates in the experimental lists with four ratios of identical vs. non-identical cognates (50-50; 25-75; 12-88, and 0-100, respectively). Results showed that the CFE gradually decreases as the proportion of identical cognates also decreases. These findings cannot be explained by mechanisms of response competition, but instead seem to imply a dynamic and language-specific top-down regulation of lexical activation.