A total of 96 patients with respiratory failure who required mechanical ventilation admitted to the PICU, Chulalongkorn Hospital from July 1998 to June 1999 were reviewed to evaluate the PRISM score for mortality prediction and to identify factors that might influence the outcome. The statistical difference in outcome between the 2 groups (survivors and non-survivors) were underlying diseases, age, maximum positive inspiratory pressure (PIP), maximum positive end expiratory pressure (PEEP), maximum fractional inspiratory oxygen (FiO2) and PRISM score (p < 0.05). However, based on the original logistic regression equation, the predicted mortality from PRISM score in our study was much lower than our actual mortality (2.4% vs 26.0%). The sensitivity and specificity of mortality predicted by PRISM score calculated at cut-off r = 0.0 (expected mortality = 50%) was 4 per cent and 97 per cent respectively. In conclusion, the original PRISM score underpredicted the mortality outcome in our patients with respiratory failure. This suggests that PRISM score is population dependent and should be modified before being used with our patients.