This study aims at assessing the specific effects of bidirectional filters (BF) and spectral filters (SF) on signal-averaged ECG (SAECG) analysis. The GISSI-3 Arrhythmias Substudy collected SAECGs of 598 patients 10 +/- 4 days after myocardial infarction (MI) from 20 Italian coronary care units. BF and SF were applied on 340 and 258 patients, respectively. QRS duration (QRSD), low amplitude signal duration (LAS40), and root mean-square-voltage (RMS40) were measured with filters set at 40 to 250 Hz. For ventricular late potentials (VLP) detection filter-specific criteria were adopted: QRSD > 114 ms, LAS40 > 38 ms, RMS40 < 20 microV for BF and QRSD > 120 ms, LAS40 > 38 ms, RMS40 < 20 microV for SF. VLP were considered present if any 2 of the criteria were met. The QRSD obtained by BF (100.6 +/- 13 ms) was shorter (P < .0001) than that obtained by SF (109.1 +/- 12 ms). Nevertheless, a higher prevalence of VLP for patients with BF than for patients with SF was found (23.8% vs 16.7%; P < .04). Indeed, filter-specific criteria were able to avoid any differences in the prevalence of abnormal QRSD and LAS40, but not of RMS40 (25.6% vs 17.1%, P < .02). Finally, the difference of VLP prevalence was mainly owing to the higher number of abnormal pairs of RMS40 + LAS40 (58% vs 44%) for BF than for SF. This multicentric study suggests that after MI, BF and SF produce discordant results on low-amplitude signals of filtered QRS that are not avoided by adopting filter-specific criteria. On the contrary, specific criteria seem to be suitable for comparison of QRSD between different SAECG devices in post-MI patients.