A method is proposed which significantly reduces the artefacts commonly experienced in dual radionuclide subtraction studies. Images of two radionuclides recorded simultaneously differ in resolution, sensitivity and attenuation. Also, one image will include scatter from the second higher-energy radionuclide. As a result severe artefacts are likely to occur when the two images are subtracted. In order to minimise the depth dependence of resolution, attenuation and scatter, the geometric mean of conjugate views was considered. From experimental work with activity placed in a depth of water it was demonstrated that the number and spatial distribution of scattered photons recorded in any energy window could be accurately predicted from the geometric mean image recorded in the photopeak. This prediction was accurate, independent of the depth of the source in water for a range of phantom dimensions. Differences in the instrument sensitivity and resolution at different energies can also be readily compensated for by using geometric mean images, as can differences due to the variation in attenuation. In practice three factors can be experimentally determined for any pair of radionuclides: a scatter ratio, a scatter function and a resolution compensation function. These data are then used to improve the dual-radionuclide subtraction analysis. The ability of the technique to significantly reduce subtraction artefacts has been demonstrated in phantom studies.