To characterize diastolic function from transmitral Doppler data, the image's maximum velocity envelope (MVE) is fit by a model for flow velocity. To reduce the physiologic beat-to-beat variability of best-fit determined model parameters, averaging of multiple cardiac cycles is indicated. To assess variability mathematically, we modeled physiologic noise as a random (normally-distributed) process and evaluated three methods of averaging (1, averaging model parameters from single images; 2, averaging images; and 3, averaging MVEs) using clinical datasets (50 continuous beats from 5 subjects). Method 2 generates a positive bias because low-velocity beats will not contribute to the composite MVE. The difference between Methods 3 and 1 is less than 2.0 E-5 (m/s)2 for uncorrelated model parameters. Input having 10% beat-to-beat variation yields a bias of <4% for model parameter mean. Hence, Method 1 was, in general, more robust than Method 3.