In the digital age, recommendation systems personalize online content based on user preferences, enhancing satisfaction and loyalty. Understanding the factors that influence individual preferences is therefore crucial. While factors like repeated exposure, emotional states, social context, and memory can affect preferences, the neural mechanisms leading to divergent preferences from similar experiences remain underexplored. In our recent study, we investigated how subjective experiences are transformed into preferences, addressing the challenge of disentangling individual differences in subjective experiences from differences in transformation processes. To achieve this, we employed a naturalistic neuroimaging paradigm. Participants underwent a two-phase study: first, they watched diverse videos inside an fMRI scanner without knowledge of subsequent tasks, ensuring unbiased neural responses. In the second phase, they expressed preferences between video pairs through a binary forced-choice task and provided subjective appraisals for each video across various dimensions. Using machine learning methods, our findings showed that within-subject multivariate brain models could predict individuals’ preferences with high accuracy. Further intersubject representational similarity analysis revealed a strong association between similarities in within-subject brain models and appraisal models. By combining naturalistic neuroimaging paradigms with advanced machine learning techniques, we offer a new approach to disentangling subjective experiences from decision-making behaviors. This research advances computational cognitive neuroscience and has practical implications for developing more effective personalized recommendation systems that account for individual uniqueness in preferences and choices.