A detailed assessment of intra- and inter-reader variation in the interpretation of brain SPECT scans has been performed. A random sample was selected from scans performed at a community/teaching hospital in Seattle. Scans were interpreted independently by three experienced readers who were blinded to all patient information. Forty-eight scans were interpreted twice by each reader, for a total of 288 readings. Readers recorded detailed assessments of individual lesions and overall impressions using a standardized reporting form. Intra-observer agreement as reflected in per cent agreement for severity scores ranged from 65% to 100%. Intra-observer agreement on the 'overall impression' was very good for Alzheimer's pattern (kappa=0.73-1.00), and fair to good for the 'heterogeneous pattern' (kappa=0.30-0.63). Inter-observer agreement, as reflected in per cent agreement, ranged from 29% to 100%. Inter-observer agreement about the 'overall impression' was fair to moderate for Alzheimer's pattern (kappa=0.24-0.54) and was poor for the descriptors 'heterogeneous' and 'normal'. It is concluded that brain SPECT has great potential value in many important conditions. This study demonstrates a need for further work in the areas of pattern definition and reduction of observer variation.