<p>The Eötvös correction is a major factor affecting the accuracy of marine gravity survey. To mitigate the Eötvös correction error in irregular nonlinear navigation, we proposed a new Eötvös correction algorithm that emphasizes the optimization of velocity and course calculation intervals. Unlike traditional algorithms that solely rely on positional information, the new algorithm establishes a real-time correlation between the vehicle’s navigation state and position information through circular fitting. By comprehensively incorporating both navigation state and positional data, it optimizes the calculation of velocity and course, thereby enhancing the effectiveness of Eötvös correction while mitigating the impact from Global Navigation Satellite System (GNSS) position errors. Utilizing marine gravity measurements surveyed in the South China Sea, the root-mean-square error (RMSE) of discrepancies at gravity cross points processed by the new algorithm was 1.65 milli gals (mGal), outperforming other methods. Furthermore, as position errors escalate, it maintains high accuracy, with the RMSE staying within 2.33 mGal. In contrast, the results of the differential GNSS method exhibit significant fluctuations. While the track fitting method producing smooth results, it introduced distortion in the Eötvös correction due to over-smoothing. The results demonstrate that the proposed algorithm enhances the effectiveness and robustness of Eötvös correction, and is well-suited for marine gravity surveys conducted using irregular nonlinear navigation patterns, where course and velocity undergo real-time variations.</p>