This paper proposes an enhanced augmented radial basis function (eARBF) metamodel combining radial basis function and polynomial chaos expansions for global sensitivity analysis. First, the conditions required for the augmented radial basis function are constructed from the perspective of variance decomposition, after which the analytical expressions of the Sobol’ indices are deduced. To improve the generalization performance of ARBF, an anisotropic technique is proposed based on the local density of sample points. Furthermore, the recursive evolution LHD and efficient K-fold cross-validation method are adopted to reduce computational efforts. Five cases are presented to demonstrate the performance of the proposed approach. In all cases, eARBF yields satisfactory results with lower computational effort. The results indicate that the approach proposed in this paper is promising for GSA of engineering problems.