A magnetic inverse solution that uses a single current dipole in a homogeneous volume conductor with realistic torso shape was tested numerically to establish the effect of magnetic noise, number of measurement points, and torso size on the localization accuracy. Seven different sites of cardiological interest were selected as locations for the source dipole. The three components of the magnetic field were calculated as if measured by second order gradiometers, Gaussian noise was added, and Monte Carlo tests performed for inverse solutions using a single field component, or all three combined. It was found that for any of the single component solutions, and a signal-to-noise ratio of 100, 25 measuring points are sufficient for good accuracy; just 12 points are needed if all three components are used together. If, however, the torso size of the inverse solution is different from that of the field data by 10 or 20%, a larger error occurs, even for 56 measurement points and no noise. In this case, the field component orthogonal to the measurement grid, Bz, yields better results than the other two components, or even all three combined. We conclude that a multichannel system measuring the z component of the magnetic field in about 30 locations would be the best choice to locate a dipolar source, provided the torso of the field data is closely matched by the model used in the inverse solution. To this effect, scaling of the torso model can easily be included in the computation. Imaging techniques could be used to accommodate different torso shapes.