A proportion of patients with computed tomographic (CT) scan appearances of malignant brain tumour undergo conservative management, despite the absence of histological confirmation of the diagnosis. Concern that this policy risked misdiagnosing a benign tumour prompted us to examine the accuracy of CT scanning in diagnosing malignant lesions. The study was designed to determine whether within a group of 300 patients with intracerebral mass lesions of known pathology, two sub-groups existed: one with appearances so specific for malignant glioma that biopsy was unnecessary, and the other in which the appearances were characteristic of malignancy, though not specific for glioma. Three neuroradiologists independently reviewed the CT scans, together with brief clinical details. When diagnosing malignant tumours, all made errors: nine benign lesions were considered to be malignant. When diagnosing malignant glioma, one neuroradiologist made errors, but the other two adopted a more cautious approach and were accurate. The restricted a "certain" diagnosis to about one in five scans considered to show malignant tumour. Those diagnosed specifically as malignant glioma were intrinsic, irregular, mixed density lesions, exhibiting variable enhancement and infiltrating the peri-ventricular tissues, especially the corpus callosum. Using these criteria, they could correctly identify a small proportion of patients with malignant gliomas. In all other patients, biopsy remains the only means of obtaining a definitive diagnosis. [ABSTRACT FROM AUTHOR]