Several post-processing methods were proposed to overcome the excess noise factor induced by EMCCD multiplication register. Each method has a unique effect on SNR. However, since SNR does not account for photometric accuracy, it cannot be reliably used to directly compare the performance of these algorithms. A normalized quadratic error that accounts for both SNR and accuracy is proposed as an alternative figure of merit. This approach provides a quantitative and rigorous comparison. Using both experimental and simulated frames in the faint-flux range, it is used to compare the existing EMCCD post-processing methods.