Late-onset sepsis (LOS) (i.e., sepsis in a neonate after 72 hours of life) is associated with high mortality and significantly prolonged antibiotic exposure and hospital stay in neonates admitted to intensive care units (ICU). In this study, we assessed the reliability of serum C-reactive protein (CRP) as a determinant of antimicrobial treatment duration of LOS. From January 1996 to December 2002, all consecutive infants aged <28 days admitted to a single medical-surgical ICU and diagnosed with primary LOS were enrolled in a prospective, intervention trial with historical controls. Only blood culture-positive LOSs were included. Exclusion criteria were: age >28 days at diagnosis of LOS, development of site-specific infection, and central venous catheter-related LOS. From January 1996 to July 1998 (historical control group), antimicrobial treatment of LOS was offered for at least 14 days. From August 1998 to December 2002 (intervention group), neonates underwent serial semiquantitative measurements of serum CRP, and antimicrobial treatment was discontinued when CRP was <12 mg/L. Primary efficacy endpoint was the duration of antimicrobial therapy. Secondary efficacy endpoints were the proportion of relapsing sepsis within 72 hours of antibiotic withdrawal and the overall mortality rate. The historical control group comprised 76 neonates developing 85 episodes of LOS; 138 LOS occurring in 120 patients comprised the intervention group. Length of antimicrobial treatment of LOS was significantly shorter during the second study period (16 days vs. 9 days, p<0.001). Secondary efficacy endpoints showed similar rates of relapsing sepsis and overall mortality in both time periods.