Objective: To quantify variation of mean paediatric length of stay (LoS) among intensive care units (ICUs) in Australia and New Zealand. Method: Retrospective data from Australian and New Zealand institutions that admitted children to ICU were analysed. The data were collected between 1997 and 2006, providing a total of 123 institution years of data (an average of 6.15 years per site). Using 47 068 admissions, LoS was modelled as the outcome variable of a G regression with a child's risk factors entered as fixed effects (allowing adjustment for case mix) and variation among ICUs modelled using a random effect. Results: Six Australasian ICUs had an average riskadjusted LoS for children, which was significantly shorter than average, whereas five had an average LoS that was significantly longer than average. The remaining nine sites had average LoS that were not significantly different from the average (at the 95% level). Among other risk factors, previous admission to an ICU and respiratory support within the first hour of admission were both associated with prolonged LoS. Conclusion: There was significant variation in paediatric LoS at the ICU level not accounted for by patient casemix. This has important implications for efficiency of ICU processes and, possibly, other components of quality of patient care in those institutions with longer LoS.