A three-dimensional reconstruction method of a small-scale magnetic target is studied in this paper. Firstly, the tilt angle and local wavenumber are calculated from the measured magnetic gradient tensor data, leading to estimates of an approximate horizontal boundary and central depth of the magnetic target. From these, a spatial range of the inversion space and depth weighting function of the magnetic target are determined. Secondly, the inversion space is divided into cuboid cells, and the inversion equations for the magnetic parameters are established. They are regularized subject to the determined spatial range and depth weighting. Finally, a generalized cross validation method is used to obtain optimal regularization parameters, allowing the magnetization intensity of each cuboid to be estimated, and the three-dimensional profile of magnetic target to be delimited. Our simulation and experimental results show that the proposed method can realize, with a high degree of fidelity, the three-dimensional reconstruction of a simple small-scale magnetic target.