To assess interobserver variability in estimation of brain atrophy based on CT, four neuroradiologists examined CT brain images of 150 consecutive patients without focal lesions. An independent neuroradiologist made the following quantitative measurements: frontal horn index, subarachnoid space area and the ratio between subarachnoid space area and inner skull space area. Level of agreement was fair for the presence (k = 0.24), slight for the degree (mild, moderate, severe) (k = 0.24) and moderate for the type (cortical, subcortical, mixed) of atrophy (k = 0.59). There was a highly significant correlation between the number of observers agreeing and quantitative measurements. We concluded that neuroradiologists' subjective estimation of brain atrophy alone is not reliable. Quantitative measurements would be needed in cases where the presence of brain atrophy might determine clinical decisions.