We performed a prospective observational (noninterventional) study of hypothermia blanket use in a population of adult intensive care unit patients with body temperatures of > or = 102.5 degrees F. Thirty-nine of ninety-four febrile episodes (in 83 patients) were treated with hypothermia blankets. Logistic regression revealed that the strongest independent predictors of hypothermia blanket use were a temperature of > or = 103.5 degrees F (odds ratio [OR] = 17), mechanical ventilation (OR = 25), and acute central nervous system illness (OR = 7.5). Hospitalization in the medical intensive care unit was strongly associated with avoidance of this therapy (OR = 0.023). Treatment with a hypothermia blanket was ordered by a physician in only 15% of cases. The mean cooling rate was the same (0.028 degree F/h) for blanket-treated and control patients. Multivariate Cox regression and factorial and repeated measures of analysis of variance revealed that blanket treatment was not more effective than other cooling methods. However, this treatment was associated with more "zigzag" temperature fluctuations of > or = 3 degrees F (56% of blanket-treated patients vs. 18% of control patients; P < .001) and rebound hypothermia (18% vs. 0; P = .001). Hypothermia blanket therapy is primarily a nursing decision. We conclude that in addition to being no more effective than other cooling measures, hypothermia blanket therapy was associated with more temperature fluctuations and with more episodes of rebound hypothermia.