A study was performed to investigate the morbidity of mediastinal lymph node dissection (MND) and to establish its contribution to the accuracy of staging in surgically treated non-small cell lung cancer. Between 1988 and the middle of 1991 a systematic sampling of mediastinal lymph nodes was done in 20 patients and a MND was carried out in 65 patients. Data from these patients were compared with those from a control group of 70 patients operated on in 1986 and 1987, who would have had MND if they had been treated in the years after 1988. The groups were comparable according to important clinical characteristics. There was a significantly greater fluid production via the drains in the groups with systematic sampling and MND, compared with the controls. Volume of blood lost during the operation and number of units blood transfused perioperatively were not significantly different between the groups. Three lesions of the recurrent laryngeal nerve and two episodes of chylothorax were observed, all probably caused by MND. The discovery ratio for N2 disease in the MND and systematic sampling groups together compared with the control group was 2.1, with a 95% confidence interval from 1.04 to 4.2.