Potential field data plays a vital role in geophysical exploration especially for mineral and hydrocarbons. Inversion of such static fields, like gravity and magnetic is a proper approach to delineate the subsurface distribution of the physical property. In ore deposits prospection, studying the subsurface distributions of densities is of critical importance. A proper strategy is to subdivide the subsurface 2D model space into rectangular mesh of prisms of different densities. The problem is highly ill-posed. To solve for the unknown densities, a new hybrid scheme, using Bat algorithm (BA) with the preconditioned conjugate gradient (PCG) technique, has been applied. The scheme benefits from the advantages of both techniques. The bat algorithm acts as a rapid build-up of the model, while PCG adapts the approximated subsurface model. The new approach is applied to a free-noise synthetic data and then to data with different levels and kinds of noise. Moreover, the technique is compared with other inversion methods (stochastic and deterministic ones) to test the efficiency and efficacy. Satisfying results are achieved through the new scheme. However, the sensitivity of the problem has been studied and presented for more understanding and testing for the inverted model. The applicability and validity of this algorithm are further verified by applying it to real residual gravity anomalies in southern California, USA, and Mexico-USA border. Also, two successful applications are done, for mineral prospection in China. Results are in good agreement with those obtained by published literature.