The clinical usefulness of two-dimensional measurements of regional cerebral blood flow (rCBF) for diagnosis of dementia is discussed on the basis of data from 150 patients with autopsy-based diagnosis. The specificity of Alzheimer's disease is found to be very good but the sensitivity only in the order of 75-80%, the majority of the mislabelled cases having selective incomplete white matter infarcts in addition to their Alzheimer pathology. Frontal lobe dementia is recognized by rCBF with high sensitivity but fairly low specificity due to the fact that frontal flow decreases are shared with several other disorders. Regarding cerebrovascular dementia the sensitivity is rather good (95%) while specificity is somewhat lower (70-75%). It is argued that rCBF should be interpreted diagnostically only when clear-cut findings are made preferably in a series of measurements. The noninvasiveness, repeatability, low radiation dose and low cost makes the present method an excellent tool for research as well as for clinical routine use.