Existing research shows that identical cognates are read more quickly than noncognates. However, most studies focused on words presented in isolation or embedded in sentences. To address this gap, our exploratory eye-tracking study is the first to investigate the processing of cognates and noncognates in English subtitles. First, we tested whether cognates differ from noncognates in terms of word processing. Second, we explored whether gradual changes in the amount of cross-linguistic overlap predict cognate processing (and potentially learning). We recorded the eye movements of 35 L1-Polish adult participants with high L2-English proficiency while they watched videos with English subtitles displayed at three different speeds. The mixed-model analyses showed that cognates and noncognates are processed for longer in slower subtitles than in faster subtitles. Also, we found no difference in processing between cognates and noncognates. However, more similar cognates were processed longer than less similar cognates, except for identical cognates (e.g., Polish/English "minus"), which were processed the fastest. The discussion addresses several implications for L2 lexical learning via audiovisual materials and makes some recommendations for future research.