OBJECTIVE: Surface electromyography (EMG) decomposition is crucial for identifying motor neuron activities by analyzing muscle-generated electrical signals. This study aims to develop and validate a novel motor unit action potential (MUAP)-based method for surface EMG decomposition, addressing the limitations of traditional blind source separation (BSS)-based techniques in computation complexity and motor unit (MU) tracking. METHODS: Within the framework of the convolution kernel compensation algorithm, we developed a MUAP-based decomposition algorithm by reconstructing the MU filters from MUAPs and evaluated its performance using both simulated and experimental datasets. A systematic analysis was conducted on various factors affecting decomposition performance, including MU filter reconstruction methods, EMG covariance matrices, MUAP extraction techniques, and extending factors. The proposed method was subsequently compared to representative BSS-based techniques, such as convolution kernel compensation. MAIN RESULTS: The MUAP-based method significantly outperformed traditional BSS-based techniques in identifying more MUs and achieving better accuracy, particularly under noisy conditions. It demonstrated superior performance with increased signal complexity and effectively tracked motor units consistently across decompositions. In addition, directly applying the MU filters reconstructed from MUAPs to decomposition exhibited marked computational efficiency. CONCLUSION AND SIGNIFICANCE: The MUAP-based method enhances EMG decomposition accuracy, robustness, and efficiency, offering reliable motor unit tracking and real-time processing capabilities. These advancements highlight its potential for clinical diagnostics and neurorehabilitation, representing a promising step forward in non-invasive motor neuron analysis.