Four hundred and forty-five randomly selected hospitalised patients had their nutritional status assessed from anthropometric, haematological and biochemical data. Nutritional status was compared between survivors and non-survivors at baseline, 6 weeks and 6 months. Using Cox's proportional hazard analysis, we measured the association between nutritional assessment variables and 1-year mortality after adjusting for disability, chronic illness, medications, smoking and tissue inflammation. Nutritional status was significantly worse amongst non-survivors compared with survivors, and non-survivors showed marked and significant deterioration in all measures of nutritional status compared with survivors. After adjusting for poor prognostic indicators the hazard ratios of death in the fourth, third and second quarters of both baseline serum albumin and mid-upper arm circumference distributions relative to the first were 0·68, 0·77 and 0·58 (trend P = 0·013) and 0·61, 1·0 and 0·87 (trend P = 0·005) respectively. Intervention studies are needed to determine whether the relationship between malnutrition and the poor outcome highlighted by the present study is causal or a mere association.