This study concerns 45 patients operated on for a primary bronchial cancer and without local or regional extension on a standard pulmonary radiograph. All subjects had a computed tomographic examination (TDM) on average 28 days before thoracotomy. The comparison was established and the possibility of excising the tumour, joined to an anatomopathological study of the structures removed. The series included only those patients with the following minimal conditions: tumour volume of 4 cms, central tumour, or close to the chest wall. The degree of pleural extension was predicted with a sensibility of 92%, a specificity of 72% (accuracy of 78%). The parietal extension was predicted with a sensibility of 63%, a specificity of 100% (accuracy of 95%). Direct invasion of the mediastinum, present 16 times, was recognized by the scanner in 8 occasions (sensibility 50%) and excluded 23 times out of 25 (specificity 85%). Extension to mediastinal ganglions was detected by the scanner 10 times out of 15; the absence of invasion 27 times out of 30. Thus, if sometimes the TDM examination showed the certainty of local or regional tumour extension, in many cases it did not by itself allow this conclusion. This was true for tumours flush with the pleura or mediastinum because their resolution was insufficient to distinguish neoplastic tissue from normal or inflammatory tissue. The TDM always established a remarkable "map" of the mediastinal glands guiding the biopsy or the thoracotomy in case of enlarged glands. One of the limits includes the possibility of occult extension without glandular hypertrophy.