<p>Direct current (DC) resistivity and seismic refraction methods are widely used for near-surface structure exploration. Joint inversion of these two methods can reduce ambiguity and improve inversion results. Traditional joint inversion algorithms may strongly depend on the direct relationship between rock physical properties or structural similarity. To make joint inversion more flexible, this paper introduces a novel joint inversion method constrained by Gaussian mixture clustering. By incorporating the Gaussian Mixture Model (GMM) into the objective function to mutually constrain resistivity and velocity, and replacing the statistical relationship of rock physical properties with the GMM probability density function, we develop a two-dimensional joint inversion algorithm for DC resistivity and seismic refraction methods. Synthetic examples using rectangular and step models show that GMM can be applied not only to single-parameter constrained inversion but also to joint inversion. When accurate statistical relationships of rock physical properties are used as prior information, GMM joint inversion yields the best results, significantly outperforming single inversion and cross-gradient joint inversion. Even when inaccurate prior information is used for GMM joint inversion, the inversion results can still effectively recover the shape and position of the anomalies, indicating that the algorithm does not entirely rely on rock prior information and has promising application prospects.</p>