Introduction: The purpose of the present study was to critically review the existing body of evidence on ventilation modes for infants and children up to the age of 18 years. Methods: The PubMed and EMBASE databases were searched using the search terms 'artificial respiration', 'instrumentation', 'device', 'devices', 'mode', and 'modes'. The review included only studies comparing two ventilation modes in a randomized controlled study and reporting one of the following outcome measures: length of ventilation (LOV), oxygenation, mortality, chronic lung disease and weaning. We quantitatively pooled the results of trials where suitable. Results: Five trials met the inclusion criteria. They addressed six different ventilation modes in 421 children: high-frequency oscillation (HFO), pressure control (PC), pressure support (PS), volume support (VS), volume diffusive respirator (VDR) and biphasic positive airway pressure. Overall there were no significant differences in LOV and mortality or survival rate associated with the different ventilation modes. Two trials compared HFO versus conventional ventilation. In the pooled analysis, the mortality rate did not differ between these modes (odds ratio = 0.83, 95% confidence interval = 0.30 to 1.91). High-frequency ventilation (HFO and VDR) was associated with a better oxygenation after 72 hours than was conventional ventilation. One study found a significantly higher PaO(2)/FiO(2) ratio with the use of VDR versus PC ventilation in children with burns. Weaning was studied in 182 children assigned to either a PS protocol, a VS protocol or no protocol. Most children could be weaned within 2 days and the weaning time did not significantly differ between the groups. Conclusions: The literature provides scarce data for the best ventilation mode in critically ill children beyond the newborn period. There is no evidence, however, that high-frequency ventilation reduced mortality and LOV. Longer-term outcome measures such as pulmonary function, neurocognitive development, and cost-effectiveness should be considered in future studies.