The dependence of action potential duration (APD) on the preceding diastolic interval (DI), i.e., restitution, has been purported to predict the development of alternans and reentrant arrhythmias. However, restitution depends on the history of activation (i.e., memory), and its relevance to arrhythmia induction and maintenance is unknown. Using a dual-camera video imaging system, we recorded action potentials from thousands of sites on the surface of the isolated pig heart. A steady-state pacing (SSP) protocol was performed to generate the SSP APD restitution curve. During SSP, the minimum DI and APD were 57 +/- 6 ms and 107 +/- 6 ms, respectively. The restitution slope was >1 for DIs <85 +/- 5 ms; however, alternans were not observed. Abrupt decreases in cycle length (CL) resulted in a rapid (<5 beats) decrease in APD followed by a slower decrease to "steady state." DI, APD pairs for the initial beats following these rate changes were significantly above the SSP restitution curve. DI, APD pairs measured during sustained ventricular fibrillation clustered significantly below the SSP restitution curve, at significantly shorter APDs (57 +/- 4 ms) and DIs (49 +/- 6 ms) than could be achieved during SSP. In addition, abrupt increases in CL following SSP resulted in APDs significantly shorter than those predicted from the SSP restitution curve. Our results indicate that the responses of APD and DI to sudden rate changes and during arrhythmias are not predicted by the SSP restitution relationship. Acute dynamics act to damp out the proarrhythmic oscillations predicted from the SSP restitution curve.