Objective To construct a multidimensional and quantifiable evaluation system for rehabilitation Chinese medicine tuina techniques, and to explore a pattern recognition method based on surface electromyography (sEMG) and kinematic parameters, thereby facilitating the automated and objective grading of tuina practitioners’ technique skills. Methods This cross-sectional observational study recruited sixty right-handed male tuina practitioners and students from the rehabilitation and tuina departments of three tertiary hospitals in Wuhan, Hubei Province (including the Affiliated Hospital of Hubei University of Chinese Medicine) between June 2024 and December 2025. Based on predefined criteria of clinical experience and annual operation hours, the subjects were divided into an experienced group, a skilled group, and a novice group, with 20 subjects in each group. Their upper limb 8-channel sEMG signals and kinematic data of the wrist and elbow joints during standardized“ rolling technique” were synchronously collected using a wireless sEMG system and an optical motion capture system. Various feature parameters, including time domain, frequency domain, and nonlinear dynamics, were extracted through intelligent methods and statistically compared. In data processing, principal component analysis (PCA) was employed for feature dimensionality reduction of parameters. Support vector machine (SVM), random forest, and deep learning models were applied to subject classification and recognition, respectively, and the classification performance of these three models was compared. Results Significant differences were observed in the kinematic and electromyographic feature test results among the three groups of subjects (all P＜0.05). The cumulative contribution rate of the PCA method reached 85.2%, the accuracy of the random forest model in classification tasks was as high as 94.7%, and the accuracy of the deep learning model in distinguishing between experienced and novice groups was as high as 92.8%. Conclusion The method of integrating multi-channel sEMG time-frequency-nonlinear features can play an important role in the assessment of tuina techniques, enabling an objective evaluation of tuina techniques and providing a theoretical basis for subsequent tuina technique assessment and intelligent clinical quantitative evaluation.