The aim of this study was to examine whether weight loss at presentation, in patients who were to receive chemotherapy for gastrointestinal carcinomas, influences outcome and whether nutritional intervention would be worthwhile. This study was a retrospective review of prospectively gathered data. The outcomes of patients with or without weight loss and treated for locally advanced or metastatic tumours of the oesophagus, stomach, pancreas, colon or rectum were compared. In 1555 such consecutive patients treated over a 6-year period, weight loss at presentation was reported more commonly by men than women (51 versus 44%, P = 0.01). Although patients with weight loss received lower chemotherapy doses initially, they developed more frequent and more severe dose limiting toxicity--specifically plantar-palmar syndrome (P < 0.0001) and stomatitis (P < 0.0001)--than patients without weight loss. Consequently, patients with weight loss on average received 1 month (18%) less treatment (P < 0.0001). Weight loss correlated with shorter failure-free (P < 0.0001, hazard ratio = 1.25) and overall survival (P < 0.0001, hazard ratio = 1.63), decreased response (P = 0.006), quality of life (P < 0.0001) and performance status (P < 0.0001). Patients who stopped losing weight had better overall survival (P = 0.0004). Weight loss at presentation was an independent prognostic variable (hazard ratio = 1.43). The poorer outcome from treatment in patients with weight loss appears to occur because they receive significantly less chemotherapy and develop more toxicity rather than any specifically reduced tumour responsiveness to treatment. These findings provide a rationale for attempting randomised nutritional intervention studies in these patients.