It is frequently necessary, both clinically and in the laboratory, to estimate how strong a stimulus is required to defibrillate. Current techniques for forming such estimates require the repeated induction of ventricular fibrillation (VF) and subsequent attempts at defibrillation (DF testing). DF testing can be time consuming and in the operating room may increase the patient risks. A novel scheme is presented which combines DF testing with upper limit of vulnerability (ULV) testing. ULV testing is a relatively safe procedure which yields data well correlated with defibrillation efficacy. A Bayesian statistical model of combined ULV/DF testing is presented which is both powerful and concise. The model is used in two examples to design minimum rms error protocols and estimators for the DF95 (the stimulus strength which defibrillates 95% of the time). A simulation for humans of one example solution shows that a single VF episode of combined ULV/DF testing (rms error = 23% of the mean DF95) is better than two VF episodes with DF testing alone (25%). The simulation results for a second example are directly compared with laboratory results from six pigs, showing a less than 1.0% average difference between the simulated and measured rms errors.