The upper limit of vulnerability (ULV) is the strength at or above which VF is not induced when a stimulus is delivered during the vulnerable phase of the cardiac cycle. Previous studies have demonstrated a statistically significant correlation between the ULV and the defibrillation threshold (DFT) in groups of patients. However, the correlation between ULV and DFT may not be close in individual patients. This imperfect correlation may be due to physiological factors or to limitations of the measurement methods. The reproducibility of either DFT or ULV has not been studied critically. The purpose of this study was to compare the reproducibility of clinically applicable methods for determination of DFT and ULV. We prospectively studied 25 patients with a transvenous implantable cardioverter defibrillator (Medtronic 7219D) at postoperative electrophysiological study. DFT was defined as the lowest energy that defibrillated after 10 seconds of VF. The ULV was defined as the lowest energy that did not induce VF with three shocks at 0, 20, and 40 ms before the peak of the T wave in ventricular paced rhythm at a cycle length of 500 ms. Both the DFT and the ULV were determined twice for biphasic pulses using a three-step, midpoint protocol. There was no significant difference between the two determinations of DFT (10.1 +/- 5.9 J vs 10.4 +/- 5.8 J), the two determinations of ULV (13.4 +/- 6.8 J vs 13.8 +/- 6.6) or the DFT-ULV Pearson correlation coefficients for each determination (0.84, P < 0.001 vs 0.75, P < 0.001). To analyze reproducibility, Lin concordance coefficients for second determination versus first determination were constructed for both ULV and DFT. This coefficient is similar to the Pearson correlation coefficient, but measures closeness to the line of identity rather than the line of regression. The Lin concordance coefficient for ULV was higher than that for DFT (0.93, 95% CI 0.85-0.97 vs 0.64, 95% CI 0.33-0.82; P < 0.01). For paired comparison of defibrillation efficacy under different experimental conditions, the sample sizes required to detect differences of 2 J, 3 J, and 4 J (80% power, P < 0.05) were 52, 24, and 15 for DFT versus 15, 8, and 6 for ULV. We conclude that a simple, clinically applicable method for determination of ULV is more reproducible than the single point DFT. Measured correlations between the ULV and single point are limited by the reproducibility of the DFT measurement.