This study was designed to test how well a commercial intervention with a highly feminized role model (Barbie) worked to improve pre-adolescent girls' interest and performance in computer science. Concurrently, this study examined how overtly feminist texts and images of real women would impact girls compared to the traditional highly feminized Barbie texts and images. Girls were randomly assigned to hear one of the four books in a 2 (pictures: Barbie illustrations or photos of real women) × 2 (text: original or feminist) design. Sixty-eight 6-13-year-old girls (M[subscript age] = 9.48, SD = 1.41) were read a book, reported their interest in computer science, created an avatar for themselves, received coding training, and completed a coding assessment. It was predicted that girls who had seen a book with a highly feminized model and heard the original text would choose to spend more time on appearance-based skills (creating a personal avatar) over coding skills and thus would score lower on the coding assessment than girls who saw photos of real women and heard the feminist text. Results showed that the text manipulation, rather than the pictures, had the most influence on the girls' behaviors: girls who heard the feminist text performed better on the coding task and were more likely to choose a coding activity over an avatar creation activity. The girls' individual characteristics (gender attitudes and gendered avatar choices) showed significant associations with their hypothetical responses to sexist comments, coding performance, and choice of coding activity. Results underscore the importance of language in the internalization of gender stereotypes.