During the past 20 years 33 patients suspected of harbouring an insulinoma have been investigated. 29 had laparotomy, and tumours were removed from 27.2 of the 29 and 1 other proved not to have an insulinoma, although preoperative imaging had suggested a tumour. Four different localisation procedures were used, and in some patients more than one technique was applied. Selective arteriography of branches of the coeliac axis showed the position of the insulinoma correctly in 9 out of 18 cases, but in all of these the tumour was felt at operation, so that the information provided was unnecessary. Arteriography gave false localisation in 4 patients and missed the tumour completely in 4 but was also negative in 1 patient not harbouring a tumour. Ultrasonic examination provided correct localisation in only 2 out of 11 instances and computer-assisted tomography in 1 out of 8. Insulin estimation in blood obtained at percutaneous transhepatic portal-venous sampling (THPVS) provided correct localisation in 2 out of 8 cases, but in only 1 of these was it needed to guide pancreatic resection. Localisation was spurious in 5 patients, and in 1 there was no evidence of a tumour at all. In 23 patients the surgeon felt and removed the insulinoma at the first operation. In 3 tumour was palpable at a second laparotomy some years later. In only 1 was no tumour felt at operation. The false-positive findings in the THPVS were caused by misinterpretation of data. For a peak of insulin concentration in the portal vein to be meaningful, it should exceed 200 mU/l and to be fully diagnostic it should be greater than 500 mU/l. Present imaging techniques are not precise enough to localise an insulinoma. An experienced surgeon has a very high probability of being able to palpate the tumour at operation, and preliminary localisation is therefore not needed in most cases.