Local identifiability was determined for a receptor-binding radiopharmacokinetic system that included measured parameters of known uncertainty. Healthy subjects and patients with severe liver disease were studied with [99mTc] galactosylneoglycoalbumin (TcNGA). Measurements during the 30-min dynamic imaging study included the count rate over liver and heart, the quantity of TcNGA injected Lo, and the fraction-of-injected dose per liter of sampled plasma f. Typical relative standard deviations for these measurements were 1, 0.2, and 5 percent, respectively. A four-state nonlinear model describing the hepatic and plasma time-activity data was then used to calculate the standard error se(pj) for model parameters representing receptor concentration [R]o, the TcNGA-receptor forward binding rate constant kb, extrahepatic plasma volume Ve, hepatic plasma volume Vh, and hepatic plasma flow F. Accounting for the measurement uncertainties of Lo and f did not significantly increase the standard errors for parameters [R]o, kb, Ve, Vh and F. When the relative errors of Lo and f were increased to 40%, the change in se(pj) ranged from 10 to 100%, with parameter Vh being the most sensitive. The exception was se(kb), the increase of which was less than 1%. Imaging studies with reduced [R]o, typically associated with patients with liver disease, resulted in greater increases in all estimated parameter errors except se(kb) which had a lower increase. Lastly, the error propagation introduced by direct measurement of the liver observational coefficients sigma 12 and sigma 13 was investigated by simulating changes in the relative standard deviation in parameters sigma 12 and sigma 13 from 0 to 40%. Imaging studies from healthy subjects showed no increase in se(pj).(ABSTRACT TRUNCATED AT 250 WORDS)