The electrical dosage of the ECT stimulus impacts on efficacy and cognitive side effects, yet seizure threshold (ST) may vary as much as 50-fold across patients. It would be desirable to predict ST on the basis of patient and treatment characteristics. In particular, concerns have been raised that benzodiazepine use and higher dosage of barbiturate anesthetics elevate ST. In a three-site study, ST was quantified at the first ECT session using an identical empirical titration procedure in 294 patients who met RDC and DSM-IIIR criteria for a major depressive episode. ST varied over a 35-fold range across patients treated with right unilateral (RUL) (n = 267) and bilateral (BL) (n = 27) ECT. Higher ST was associated with BL electrode placement (p = 0.001). Among patients treated with RUL ECT, univariate analyses indicated that higher ST was associated with advanced age (p < 0.001), male gender (p < 0.001), greater burden of medical illness (p < 0.001), weight (p < 0.01), duration of mood disorder (p < 0.01), and history of previous ECT (p < 0.05). Average lorazepam dose in the 48 hours prior to ECT was not associated with ST, but was associated with decreased seizure duration (p < 0.01). Absolute, but not weight-adjusted, methohexital dose was associated with ST (p < 0.01). Multivariate analyses in patients treated with unilateral ECT showed that only 27.6% of the variance in ST (p < 0.0001) could be predicted. In the multivariate analyses, only age (p = 0.0001), gender (p = 0.01), and methohexital dose (p = 0.0001) were independently related to ST. Low dosage of lorazepam and methohexital dosage below 1 mg/kg are unlikely to impact on ST. Given the limited capacity to predict ST, empirical titration remains the only accurate method to determine electrical dosage in RUL ECT.