The difference between the displacement of the centre of rotation (mechanical shift, MS) and the electronic centring misalignment (electronic shift, ES) in cone beam SPECT is evaluated. A method is proposed to determine both MS and ES using the centroid of a projected point source sampled over 360 degrees and the Marquardt non-linear fitting algorithm. Both shifts are characterized by two orthogonal components. This method is verified using Monte Carlo simulated point source data with different combinations of mechanical and electronic shifts. Both shifts can be determined correctly. We have also applied the proposed method to our cone beam SPECT system to determine both shifts as well as the focal length. The determined ES parameters are then used to correct the projections and the MS parameters are incorporated into a reconstruction algorithm. The point source images are reconstructed and the image resolutions with and without the shift corrections are measured. The experimental results demonstrate that the image resolution is improved after shift corrections. The experimental results also indicate that the shift parameters determined in the same experiment with the point source located at different places are consistent but change from time to time, suggesting that calibration of the system is needed on a periodic basis.