Background - Although restitution has been hypothesized to determine action potential duration (APD) during ventricular fibrillation (VF), cardiac memory may also be important. Methods and Results - Transmembrane recordings were made with a floating microelectrode from the anterior right ventricular wall in 6 pigs during up to 60 seconds of VF. The recordings were divided into 5-second intervals, and APD60 and the diastolic interval (DI) were calculated for each activation cycle throughout each interval. Stepwise linear regression was used to determine how well each APD60 [APD60(n)] was predicted by the 4 previous DIs (n-1, n-2, n-3, n-4) and the 3 previous APD 60S (n-1, n-2, n-3). A mean±SD of 3±1.5 of the variables entered the regression equation. DI(n-1) (70% of intervals) and APD60(n-1) (71% of intervals) appeared most frequently in the regression equations and were the first or second variables entered during the stepwise regression in 87% and 76% of the intervals in which they were present, respectively. The coefficients of DI(n-1) and APD60(n-1) were positive 89% and 98% of the time, respectively. R2 of the regression for all entered variables during all intervals was 0.39±0.05. Conclusions-The high incidence and positive coefficient of DI(n-1) indicate that restitution is important in determining APD during VF, whereas the similarly high incidence and positive coefficient of APD(n-1) indicate that cardiac memory is equally important. The finding that the regression equation accounts for only 39% of the variability of APD indicates that factors other than restitution and memory are also important in determining APD during VF.