We retrospectively compared the changes in serum albumin concentration and colloid osmotic pressure between survivors and nonsurvivors of prolonged (> or = 7 days) critical illness over a 2-year period from 1 July 1995. All patients had serum albumin measured daily, and colloid osmotic pressure measured 5 days a week, throughout their ICU admission. They received crystalloid and colloid infusions as well as parenteral or enteral feeding. Infusions of albumin were not used to treat hypoalbuminaemia. One hundred and forty-five patients were included, 66 nonsurvivors and 79 survivors. Nonsurvivors were significantly older than survivors [mean (95% CI): 58 (3.8) and 49 (4.1) years, respectively] and had a greater risk of death [mean (95% CI): 0.44 (0.06) and 0.28 (0.05); p < 0.05]. There was no significant difference in gender, APACHE II score [mean (95% CI): 22 (2.7) (nonsurvivors); 18 (2.3) (survivors)] or length of stay [median (interquartile range): 14 (9-27) days (nonsurvivors); 15 (9-26) days (survivors)]. There was no difference between the two groups in the absolute minimum serum albumin concentrations reached, the time to reach that minimum or the minimum in the first 7 days. However, nonsurvivors had a significantly lower mean serum albumin concentration: [mean (95% CI): 15.7 (5.1) g.l-1 compared with 18.3 (4.6) g.l-1 in survivors; p < 0.05]. They also had a lower recovery mean (the weighted mean after the minimum value): [mean (95% CI): 13.3 (5.1) g.l-1 (nonsurvivors) and 18.6 (5.3) g.l-1 (survivors); p < 0.01]. Analysis of colloid osmotic pressure results showed no difference between the groups in mean, minimum or recovery mean. Regression analysis of mean colloid osmotic pressure and albumin revealed that albumin only contributed 17% of the colloid osmotic pressure in these patients. The similar decrease in albumin in nonsurvivors and survivors may reflect the acute inflammatory response and/or haemodilution. However, survivors showed an ability to increase serum albumin concentrations, possibly owing to resumption of synthesis. The colloid osmotic pressure varied little between or within either group of patients, possibly because of the use of artificial colloids. There was no relationship between death and colloid osmotic pressure.