Aims: Despite increasingly radical surgery for oesophageal cancer many patients continue to represent with recurrent disease. This study aimed to evaluate the pattern of failure following attempted curative oesophagectomy with mediastinal and upper abdominal lymphadenectomy. METHODS: Some 212 consecutive patients undergoing R0 resection for malignancy between 1 April 1990 and 1 April 1999 were followed up for evidence of recurrence. Clinical evaluation was supported by ultrasonography, computed tomography, isotope scan, endoscopy and laparotomy with biopsy assessment if appropriate. Patients were excluded if recurrence was diagnosed on clinical grounds alone. Statistical analysis was performed using chi2 and log rank tests. RESULTS: Some 142 patients with adenocarcinoma and 70 with squamous carcinoma (SCC) were followed up for a median of 14 (range 1-108) months. Sex and age distribution were similar for both histological subtypes (men : women 3 : 1; median age 64 (30-79) years). Twenty patients died from non-cancer related causes, including 11 (5 per cent) from postoperative complications. Some 89 patients (42 per cent) developed proven recurrent disease of which seven are alive and 82 dead. The median time to recurrence was 11 (2-40) months with a median time to death thereafter of 3 (1-21) months. The pattern of recurrence was locoregional in 23 per cent (oesophageal bed 15 per cent, upper abdominal 3 per cent, upper mediastinal 3 per cent, cervical 2 per cent) and haematogenous in 18 per cent (comprising liver 8 per cent, bone 4 per cent, cerebral 3 per cent, lung 2 per cent, skin 1 per cent) with peritoneal dissemination in 1 per cent. While there was no difference in the overall pattern of dissemination for each histological subtype, the incidence of cervical and upper mediastinal recurrence was significantly higher for adenocarcinoma compared with SCC (chi2 = 5. 9, 1 d.f., P < 0.02). The timing of recurrence was similar for both histological subtypes: 60 per cent of all recurrence occurred within 12 months of surgery, with distant and locoregional recurrence occurring at a median of 10 (2-40) and 11 (2-32) months respectively. CONCLUSIONS: The low incidence of upper mediastinal and cervical recurrence suggests that more extensive lymphadenectomy is unlikely to impact upon survival. Improved staging modalities are required to identify the significant number of patients who develop early recurrence in the first year following surgery in order to offer them multimodality therapies of non-surgical palliation.