Consistency of results is extremely important in analyzing electrocardiograms (ECGs). particularly when judging critical changes between sequential records. Thus, studies in variability are very important, and those published have shown computer results to be more repeatable than those from humans. The ECG computer program of the Veterans Administration (VA) has been particularly noteworthy in this regard. However, this study elucidates a new facet, that variability is also a function of the average probability level of the diagnostic result. Minimizing previously recognized sources of variation, 3tminute ECGs were collected from 98 normal subjects. The average probability of diagnosis and the standard deviation of that average were computed and plotted in a scattergram. A curve of variation computed from the data shows that the standard deviation is nearly zero when the probability of diagnosis is high (nearly 100%) but increases to 12% when the diagnosis is in the 50% probability range. ECGs collected in a dayto-day variation study with ordinary control on error sources are compared to this curve and are compatible with this picture of variation. The conclusions consider how the probability level itself may be useful in setting criteria for sign&cant ECG changes.