The reproducibility of the parameters defining the presence of late potentials on the signal-averaged electrocardiogram is one of the limiting factors of the method. The authors studied the coefficients of correlation and reproducibility of these parameters in patients with coronary artery disease. In addition, they tried to determine which parameter was most often responsible for changing a diagnostic conclusion (i.e., presence or absence of late potentials). Two signal-averaged ECGs were recorded one after the other in 127 patients. The presence of late potentials was defined as the presence of a least two of the following criteria: total amplified and averaged QRS duration (tQRS) > 114 ms: duration of the last signal of under 40 microV (LAS) > 38 ms, and root mean square of the amplitude of the last 40 ms (RMS) < 20 microV. The correlation coefficients were 0.98, 0.96 and 0.94 for the duration of tQRS, LAS and RMS respectively (p < 0.0001). The coefficients of reproducibility were 7.0 ms. 7.0 ms and 16.1 microV respectively. Late potentials were present in 22% of patients. A change in diagnosis between the first and second recording was observed in 10 subjects (8% of the population). A combined change in LAS and RMS was responsible for 6 of these revised diagnoses, a change in LAS alone in 2 cases, of the RMS alone in 1 case and the tQRS alone in 1 case. In patients with coronary artery disease, the immediate reproducibility of the diagnosis of late potentials is affected by changes in LAS and RMS. The tQRS is only rarely responsible for a change in diagnosis. This study suggests that the result of the signal-averaged ECG should be interpreted with caution when the LAS or RMS are near their threshold values.