To apply p wave analysis of the signal-averaged electrocardiogram (SAECG) clinically, the knowledge of normal values and the verification of highly reproducible measurements are essential. In 40 healthy volunteers (ages 23-37 years), an SAECG of the p wave was performed and then repeated after 1 week and 1 month. In addition to Simson analysis (unidirectional and bidirectional filters), the data were filtered with finite impulse response (FIR) and least squares fit (LSQ) filters to obtain a better differentiation between the end of the p wave and QRS onset. All recordings were evaluated by two independent observers, and the following parameters were calculated: the duration of the p wave in the three unfiltered leads; and for the vector magnitude of the three leads x, y, and z: the duration of the p wave and the root mean square voltage of its last 30, 20, and 10 ms. The average p wave duration was significantly different among the various filter techniques: 135 +/- 7 ms (unidirectional); 97 +/- 8 ms (bidirectional); 109 +/- 8.5 ms (FIR); and 126 +/- 10 ms (LSQ) (P = 0.001 for each comparison). There was a good reproducibility of the data analyzed with the Simson method (R = 0.80-0.91). Filtering with the FIR and LSQ leads to a considerably greater variability.