Background: The equivalent dipole layer (EDL) source model for noninvasive reconstruction of cardiac electrical activity applies a nonlinear parameter estimation procedure starting from an initial estimate for activation and repolarization times. We compare two methods to determine the initial estimate for repolarization, based on the overall QRS-T polarity relation. Methods: We performed a pig experiment with simultaneous body surface mapping and in-vivo validation using epicardial electrode sock and endocardial catheters. In addition, we included a human patient with frequent premature ventricular contractions (PVCs). Initial estimate for repolarization method 1 assumes reversal of the initial estimate for activation (determined by fastest route algorithm), while method 2 assumes a linear positive relation. Results: In the pig data, we found lower errors with higher correlation coefficients between measured and reconstructed RTs using method 2 as initial estimate for repolarization for both atrially and ventricularly paced beats. This corresponds with the linear positive relation between measured activation and repolarization time and the discordance of QRS complex and T-wave polarity on the body surface potentials in both atrially and ventricularly paced beats, indicating a similar sequence of activation and repolarization. In the human data, there is a big difference in reconstructed repolarization pattern when using method 1 or 2 for initial estimate of repolarization. Here, there is a concordance in the majority of leads for the sinus beat and a discordance in the majority of leads for the PVC. Conclusion: We recommend using the relation of the QRS complex and T-wave to determine which method for initial estimate for the EDL method of the inverse solution is most suitable for each individual beat.