<p>Accurate classification of pain intensity is essential for effective pain management, yet conventional self-report measures are subjective and influenced by individual and cultural factors. This study investigates whether integrating electrocardiogram (ECG) signals with electrodermal activity (EDA) improves objective pain-intensity classification. Pain was induced via thermal grill stimulation in 23 healthy participants across three levels: low (Level 1), moderate (Level 2), and high (Level 3). EDA and ECG signals were processed using four time–frequency methods—STFT, SPWVD, CWD, and VFCDM—and provided as input to a multi-scale convolutional neural network (MS-CNN). Among the evaluated methods, VFCDM achieved the best performance for individual modalities (EDA: Accuracy 66.3%, F1-score 67.8%; ECG: Accuracy 66.7%, F1-score 66.3%) and for the combined EDA + ECG approach (Accuracy 68.5%, F1-score 70.5%). The relative performance of ECG and EDA was dependent on the time–frequency method. ECG performed better with STFT and CWD, SPWVD favored EDA, and VFCDM slightly improved EDA performance. EDA demonstrated stronger discrimination between low and high pain extremes, whereas ECG better captured high-pain responses. Importantly, integrating both modalities yielded consistently improved performance across all pain levels. These results highlight the complementary nature of EDA and ECG signals and demonstrate the potential of VFCDM-based time–frequency analysis combined with MS-CNN for robust, objective multi-level pain intensity classification. The findings suggest that carefully selected time–frequency representations, in combination with multimodal physiological signals, can enhance the reliability and precision of automated pain assessment systems.</p>