Holistic processing refers to the human tendency to perceive objects as unified wholes rather than as separate, independent parts. A key measure of holistic processing is the composite effect, where the top half of one object (e.g., a face) is combined with the bottom half of a different object, leading to the perception of a unified, novel whole—an illusion that disappears when the combined parts are laterally shifted (misaligned). Traditionally, holistic processing has been viewed as specific to faces or objects of expertise. However, recent evidence suggests that line patterns can also be processed holistically without prior training, raising the question: Do these two types of holistic processing rely on similar or distinct processing mechanisms? Here, we used deep neural networks (DNNs) to address this question. We trained three VGG16-based DNNs on object identification, face identification, or both tasks and measured the size of the composite effect using face and non-face line pattern stimuli, analogous to those used in human studies. Preliminary results revealed that holistic processing of faces occurred only in DNNs trained specifically on face identification, supporting a face-specific mechanism. In contrast, we found a composite effect for line patterns across all trained DNNs, suggesting a domain-general effect. Furthermore, holistic processing for faces emerged in later classifier layers, whereas it was evident in earlier feature extraction layers for line pattern stimuli. Our results suggest distinct mechanisms underlying holistic processing for faces and non-face stimuli in ANNs, and, we conjecture, also in human brains.