Pneumothorax in the newborn has a significant mortality and morbidity. Early diagnosis would be likely to improve the outlook. Forty-two consecutive cases of pneumothorax that developed after admission to a tertiary referral neonatal medical intensive care unit over 4 y from 1993 to 1996 were reviewed. The time of onset of the pneumothorax was determined by retrospective evaluation of the computerized trend of transcutaneous carbon dioxide (tcpCO2) and oxygen tensions. The timing of the occurrence in the notes and x-rays determined the time of clinical diagnosis noted at the time. The difference was the time the condition was undiagnosed. The overall mortality before discharge was 45% (19 cases), four patients succumbing within 2 h. The median time (range) between onset of pneumothorax and clinical diagnosis was 127 min (45-660 min). In most cases, the endotracheal tube was aspirated and the transcutaneous blood gas sensor was repositioned, and in at least 40% of the cases, the baby was reintubated before the diagnosis was made. Reference centiles were constructed for level of tcpCO2 and slope of the trended tcpCO2 over various time intervals (in minutes) from 729 infants from 23 to 42 wk gestation who needed intensive care during the first 7 d of life from the same time period. The 5-min tcpCO2 trend slopes were compared in index and matched control infants. The presence of five consecutive and overlapping 5-min slopes greater than the 90th centile showed good discrimination for a pneumothorax (area under the receiver operating characteristic curve, 89%). We concluded that 1) the clinical diagnosis of pneumothorax was late, occurring when infants decompensate; 2) trend monitoring of tcpCO2 might allow the diagnosis to be made earlier if used properly; and 3) use of reference centiles of the trended slopes of tcpCO2 might be used for automatic decision support in the future.