<p>Joint inversion of surface wave and gravity data can reduce the non-uniqueness of individual inversion and has been applied in research on the Earth’s crust and lithospheric mantle. At present, direct parameter coupling, which requires specifying a functional relationship between shear wave (S-wave) velocity and density, is primarily used; however, it can result in spurious features when the models violate the parameter relationship. Moreover, an appropriate velocity–density function is difficult to derive, and a single physical property relationship may not be suitable for all regions. We present a new joint inversion algorithm for ambient noise surface wave and gravity data, using variation of information (VI) coupling which measures the amount of information one variable contained in another variable based on information theory. The correlation between S-wave velocity and density models is established with a one-to-one relationship with VI. The effectiveness of the algorithm is verified using synthetic and field data. The synthetic data analysis results indicate that density anomalies are accurately captured by joint inversion, whereas they are hardly captured by individual inversion. S-wave velocity models obtained by joint inversion are more accurate than those obtained by individual inversion. The fitting parameter relationship of the joint inversion models is closer to the true model than that of individual inversion. In the field case of the southeastern Tibetan Plateau, the S-wave velocity model of joint inversion has a higher resolution than that of separate inversion. The density model obtained by joint inversion clearly reveals crustal structures, which are severely distorted in the individual inversion model. Joint inversion with VI coupling is an effective and valuable approach for inverting surface wave and gravity data.</p>