To determine risk factors for infection of hyperalimentation catheters, we prospectively studied 169 catheter systems (88 patients) by using a semiquantitative culture technique. Infection occurred in 24 (14%) catheters (16 patients), was inversely proportional to the number of previous catheters inserted by the operator (P less than .02), and was proportional to the interval between admission and catheter insertion (P less than .0005). Catheter replacement over a guidewire was no more likely to be associated with infection than was a de novo percutaneous insertion at another site (P = .6). Using a proportional hazards model, we estimated the risk of infection per day to be 1.3 times greater for a catheter if the patient had been hospitalized 50 days instead of seven days, and 3.8 times greater if the patient had a Swan-Ganz catheter at the time of insertion.