Western and Chinese traditional paintings exhibit profound cognitive and aesthetic differences: Western portraiture emphasizes perceptual realism and psychological states, while Chinese portraiture prioritizes holistic qi-yun and environment harmony. This cultural contrast raises the question of how culture shapes higher-order visual judgment and whether systematic aesthetic biases emerge. To address this, we propose a computationally grounded, experience-based approach. We first interviewed over 100 participants and used grounded theory to organize 43 aesthetic dimensions into four higher-order factors: Perceptual Realism, Expressive Power, Formal Organization, and Cultural Significance.

To avoid imposing Western standards, we employed Qwen—a Chinese-trained Large Language Model (LLM)—as a computational proxy, using this participant-derived framework to evaluate 2,000 historical portraits (1,000 Western, 1,000 Chinese). Results showed that, despite acknowledging Chinese strengths in Formal Organization and Cultural Significance, Qwen significantly favored Western works overall ($d = 0.78, p < 0.001$), particularly due to high sensitivity to chiaroscuro and anatomical detail within the Perceptual Realism dimension. This bias aligns with the cross-cultural distinction between object-focused (Western) and field-dependent (East Asian) perceptual styles. Methodologically, this demonstrates that even culturally informed LLMs can reproduce dominant aesthetic norms if the evaluation framework is implicitly biased. Theoretically, it confirms that high-level aesthetic judgment is mediated by implicit evaluative schemas. This study offers an empirically grounded pathway toward more culturally equitable paradigms in AI-assisted cognitive science.