We studied the efficiency and practicability of a new protocol for surveillance of nosocomial infection in 99 patients admitted to our intensive care unit (ICU) between October 1985 and March 1986. The protocol contained the therapy given before admission to the ICU and daily records of: (1) therapy suspected to increase the risk of nosocomial infection (ventilatory support, operations, hemofiltration, central venous and arterial catheters, Swan-Ganz catheters, etc.); (2) parameters possibly associated with bacterial infections (white blood cell count, body temperature, platelet count, creatinine clearance, hemodynamic values, clinical suspicion of infection, suspected site of infection, pathological chest X-ray, etc); and (3) bacteriologic data (results of cultures from blood, tracheal aspirate, urine, wound secretions) and antibiotic treatment. Sepsis was diagnosed in 28 patients, and 28 had positive blood cultures; these two groups were not identical. Twenty-nine patients died, more than half of them having a serious nosocomial infection. Factors associated with nosocomial infection were: fever greater than 38.5 degrees C, recurrent fever, leucocytosis, leucopenia, thrombocytopenia; duration of stay at the ICU; ventilatory support, operations, mass transfusions, and therapy with catecholamines. Use of a Swan-Ganz catheter and hemofiltration were associated with nosocomial infection, but they were usually begun after the clinical onset of sepsis. Therapy administered before admission to the ICU was not associated with nosocomial infection. Bacterial colonization of tracheal aspirate was detected in 2/3 of the ventilated patients. The most common bacteria isolated were staphylococci in blood cultures. Pseudomonas sp. and Candida albicans in tracheal secretions, and Candida albicans in urine cultures.(ABSTRACT TRUNCATED AT 250 WORDS)