Precise measurement of emotion is a key challenge in understanding human emotion and developing emotional artificial intelligence. Most existing studies have regarded participants’ emotional annotations in interval form as the ground truth of their emotional experience. However, recent studies suggest that the ordinal forms of emotional annotation better represent the emotional externalization process, offering a promising approach for precise emotion measurement. In this study, we explored the neural basis of multivariate ordinal emotion representations using a video-elicited EEG dataset (n = 123). We conducted inter-situation representational similarity analysis (RSA) and inter-subject RSA to reveal the EEG substrates of emotion variations and individual differences, respectively. Our findings indicate that both inter-situation and inter-subject variations in EEG features are better explained by ordinal emotion representations than by interval ones, supporting the ordinal nature of emotion from a neural perspective. Besides, multivariate ordinal representations showed better inter-subject reliability and higher representational similarity with EEG features compared to univariate counterparts, highlighting the co-occurrence nature of human emotions. Taken together, these findings demonstrate that multivariate ordinal emotion ratings provide a more accurate measure of emotional ground truth, which is crucial for enabling machines to precisely understand and express human emotions.