Random error is associated with breath alcohol measurements, as with al1 analytical methods. The total random uncertainty of a group of n measurements is typically determined by computing the standard deviation and requiring it to be less than some appropriate leve1 (i.e. t 0.0042 g/210 ll. The total random uncertainty has two primary sources: the instrumental method and the sample source. These are typically inseparable values. In breath alcohol testing the two primary sample sources are simulators and human hreath. The present study evaluates ten groups of simulator samples consisting of ten measurements each on BAC Verifier Datamaster instruments. The data also includes ten breath alcohol measurements from each of 21 individuals following alcohol consumption. The range of standard deviations for the simulator samples was 0.0003-0.0022 g1210 1. The range of standard deviations for the human breath samples was 0.0015-0.0089 g1210 1. Two statistics that test for homogeneity of variances were applied. The simulator samples resulted in a Cochran's C test of 0.5000 and an Fmax test of 48.9. The human breath samples resulted in a Cochran's C test of 0.1519 and an F_ test of 27.3. Al1 were significant at P < 0.001. The statistical tests demonstrated that the intragroup variability among the human subjects was comparable to the intragroup variability among the simulator samples. The data also demonstrates that the sample source (simulator or humanl is probably the largest contributor to total random uncertainty. Therefore, when duplicate breath alcohol testing from individuals shows variability in the second decimal place the cause is differences in breath samples provided and not instrumental imprecision.