Abstract: To address issues such as observation errors, model perturbations, and external disturbances in the heading control of unmanned surface vehicle (USV), a positional compact form dynamic linearization based model-free adaptive control (P-CFDL-MFAC) method is proposed. Firstly, the phenomenon of output oscillation and divergence occurring in USV heading control by the standard model-free adaptive control (MFAC) method is analyzed, identifying the fundamental cause as the mismatch between the incremental dynamic input/output (IO) generalized model of standard MFAC and such integral systems. Secondly, the P-CFDL-MFAC method is proposed by constructing a positional dynamic IO generalized model and introducing pseudo-partial derivative into the control input criterion function, and its stability is analyzed. This method addresses the inherent failure issues of standard model-free adaptive heading control systems without introducing any additional compensation term, featuring simplicity, rapidity, and robustness. Finally, simulation experiments and field experiments of USV heading control are conducted to verify the effectiveness of the method. The experimental results demonstrate that P-CFDL-MFAC method is suitable for the USV heading control system, achieving a root mean square error in heading within 3°.