The time course of intensive care for severe respiratory syncytial virus (RSV) lower respiratory tract illness may be predicted by the severity of gas exchange during the first 48 h of mechanical ventilation. To test this hypothesis, two studies were undertaken in RSV-positive mechanically ventilated patients who did not have chronic lung disease, congenital heart disease or immunodeficiency. First, a retrospective criteria-generating review of 45 infants was carried out. In these infants, more severe lower airway disease, as demonstrated by four-quadrant consolidation on chest X-ray, was associated with 'best' alveolar arterial oxygen gradients (AaDO2, torr) and mean airway pressure (MAP, cm H2O) values as follows: first 24h, AaDO2 > 400 and MAP > 10 (positive and negative predictive values 100% and 97%, respectively); second 24 h, AaDO2 > or = 300 and MAP > 10 (positive and negative predictive values 91% and 100%, respectively). The second study, a prospective, hypothesis-testing, analysis of length-of-stay in 44 infants stratified according to the above AaDO2 and MAP criteria demonstrated that the duration of intensive care was longer in the severe group: median (interquartile range in days) 17 (15-39) vs 7 (4-8) (p < 0.01). We suggest that, in mechanically ventilated infants with RSV, the time course of intensive care is predictable based on early clinical features and respiratory parameters. Therefore reports on the effectiveness of special therapies using intensive care stay as a measure of outcome should be interpreted with respect to these observations before drawing conclusions about efficacy.