In a prospective blind study 380 daily serum samples from 55 women with preterm premature rupture of the membranes were analysed for C-reactive protein (CRP). Although the last CRP before delivery was higher in patients with histological chorioamnionitis (P = 0.007), considerable overlap between infected and non-infected pregnancies occurred, precluding the use of CRP as a diagnostic test if published normal levels were used. When upper limits were set at 30, 35, or 40 mg/l, the last CRP before delivery proved 90, 95 and 100% specific and 88, 92 and 100% positively predictive of infection in singleton pregnancies. Such high specificities are needed to prevent inappropriate intervention based on false positive results. We therefore propose upper limits for single estimations of 30, 35, or 40 mg/l depending on the relative risks of preterm delivery versus infection at various gestational ages. In addition, consecutive values greater than 20 mg/l appeared highly predictive of infection.