Biophysical atrial simulation can improve therapies by simulating differ-ent ablative and pharmacological strategies, although their use in clinical practice is limited by their high computational cost. Simpler cardiac automa-ta can achieve acceptable timeframes, calculating the action potential duration (APD) from the previous diastolic interval (DI), although it is necessary to question whether this approach is sufficient for short- and long-term simula-tions. Biophysical simulations (Elvira software) were carried out on a rectangular atrial tissue (0.3x20x0.025 cm, 2106 cells, A) using the Coutermanche model. Eleven long-term S1-S2 pacing protocols with 45 sets of 15xS1+S2 activations were simulated, with both increasing and decreasing S2 (100ms to 1000ms in 20ms steps, B) and S1 intervals (300ms to 700ms in 40ms steps). APD at 90% of the amplitude at current (APD+1) and previous (APD0) activation, and DI as time from 90% amplitude to next repolarization, were measured. Analysis of 7920 simulated activations showed an expected increase of the APD+1 with the previous DI interval (C). Influence of short-term memory at long-term simulations was showed as the dependency of APD+1 with the previous AP duration (APD0): shorter APD0 provoked shorter APD+1 (C), and this effect was comparable to the effect on APD+1 of previous DI. Prediction of APD+1 can be improved when both APD0 and DI are used (error of 10+-14 ms, D), compared to using DI alone (23+-21 ms, p<<0.001, D). Atrial automata should consider short term memory, as duration of previ-ous activations, to accurately estimate posterior APD in long-term simula-tions, in order to mimic the natural electrophysiological response.