Computer-generated and real curves were used to study the random error in the blood flow estimates calculated from xenon-133 cerebral clearance curves. The initial count rate affects the scattering under a threshold of 5,000 counts/s, but above this figure its influence is negligible. Flow values, relative weight of gray matter, and processes of analysis have also an effect. For the stochastic analysis, a theoretical standard deviation calculable from each experimental curve was proved. For other methods, empirical formulas were established. Using these formulas, intervals of uncertainty can be calculated allowing a useful comparison of the results obtained on the same patient, from different areas or from subsequent tests. Such intervals also allow the comparison of mean flow values obtained from a particular curve by the three usual processes of analysis. With high initial count rates, more than 75% of the intervals of uncertainty are disjointed. This suggests that the random error plays little part in the differences observed.