The administration of a known concentration of oxygen is an important part of routine care of the sick patient. Many devices are currently available. The actual concentration of oxygen that can be delivered by these devices can be affected by several factors, both from the patient as well as the device itself. Measuring the F(i)o(2) delivered to the lungs in vivo can be both difficult and potentially uncomfortable for the subjects. We constructed a model using a resuscitation manikin, a ventilator and a set of bellows to simulate ventilation. With this model we tested a series of devices - variable performance, fixed performance and high flow - at two fixed tidal volumes. The respiratory rate was increased and its effect on the oxygen concentration assessed. Variable performance systems such as the Hudson mask deliver a significantly reduced oxygen concentration at high respiratory rates. Fixed performance systems delivering 24-40% oxygen deliver appropriate oxygen concentrations across the range of respiratory rates, whereas those delivering 60% show a reduction in performance. High flow systems show no failure of performance at increased respiratory rates.