Aims: Our purpose is to assess different methods for solving the electrocardiography imaging (ECGI) inverse problem based on different regularization methods but also to assess their sensitivity to the nature of recorded signals. Results are provided using LV, RV and bi-ventricular pacing experiments performed in an ex-vivo pig heart, partially covered by a sock counting 108 electrodes from which only 93 are used, and suspended into a human-shaped torso tank covered with 128 electrodes recording torso potentials. Methods: We assess the zero order Tikhonov regularization in conjunction with the MFS method using different approaches for computing the optimal regularization parameter. These methods include GCV, RGCV, ADPC, U-curve and CRESO methods. In the experiment, some electrodes are too close to the epicardium so that they seem to be measuring monophasic action potential signals (MAPs) rather then extracellular epicardial potential. The electrodes suspected to measure MAPs are identified by having an electrical potential greater than a fixed threshold in the plateau phase, 250 ms after pacing. This leads to compare the computed epicardial potential to measurement results with and without considering these electrodes measurements. Results evaluation is obtained by computing the mean values of the relative error (RE) and the correlation coefficient (CC) over the QRS interval. Results: The computed mean value of the RE using the ADPC, CRESO, GCV, RGCV and U-curve are respectively 0.82, 0.84, 0.78, 0.86 and 0.98. The CC are respectively 0.62, 0.61, 0.67 0.61 and 0.60. Removing the MAPS electrodes improves the quality of the RE to 0.78, 0.81, 0.72, 0.84 and 0.97 respectively and the quality of the CC to 0.67, 0.66, 0.72, 0.66 and 0.66 respectively. Conclusion: Results show that the Generalized Cross Validation approach provides the best results in the three pacing cases and that removing the MAPs electrodes improves the quality of comparison.