In the current report we describe a successful training study aimed at improving recognition of a set of familiar face photographs in K., a 4-year-old girl with congenital prosopagnosia (CP). A detailed assessment of K.'s face-processing skills showed a deficit in structural encoding, most pronounced in the processing of facial features within the face. In addition, eye movement recordings revealed that K.'s scan paths for faces were characterized by a large percentage of fixations directed to areas outside the internal core features (i.e., eyes, nose, and mouth), in particular by poor attendance to the eye region. Following multiple baseline assessments, training focused on teaching K. to reliably recognize a set of familiar face photographs by directing visual attention to specific characteristics of the internal features of each face. The training significantly improved K.'s ability to recognize the target faces, with her performance being flawless immediately after training as well as at a follow-up assessment 1 month later. In addition, eye movement recordings following training showed a significant change in K.'s scan paths, with a significant increase in the percentage of fixations directed to the internal features, particularly the eye region. Encouragingly, not only was the change in scan paths observed for the set of familiar trained faces, but it generalized to a set of faces that was not presented during training. In addition to documenting significant training effects, our study raises the intriguing question of whether abnormal scan paths for faces may be a common factor underlying face recognition impairments in childhood CP, an issue that has not been explored so far.