The effect of the inaccuracy of the input function on CBF measured by the H2(15)O autoradiographic method was investigated. In H2(15)O autoradiography the measured input function usually includes a larger dispersion than the true input function, as well as the absolute time axis having been already lost. The time constant of the external dispersion that occurred in our continuous sampling system was evaluated as 10-12 s when the dispersion function was approximated by a monoexponential function. The internal dispersion occurring in arterial lines in a human body was evaluated as 4-6 s. Such dispersion, indispensable in a patient study, was found to produce large errors in calculating CBF, e.g., 5(10) s of the dispersion caused +15(33) and +10(20)% systematic overestimations for the 40- and 60-s accumulation time respectively. An analytical correction employing an inverse Laplace transform was applied to clinical CBF studies, and the results were compared with those from the C15O2 steady-state inhalation method. Correction by 10 s in time constant, corresponding to the external dispersion, reduced the overestimation significantly from 70-100% to approximately 20%. Further correction by 5 s, corresponding to the internal dispersion, resulted in a negligible difference (less than a few percent) from the steady-state method.