The authors constructed a 0.06 ml measurement chamber connected via high-speed valves (0.5 ms response time) to two 3 l reservoirs pressurized to 50 kPa with gases containing different concentrations of CO2. An electronic system opens the valves alternately depending on the polarity of a control voltage Ve. Two walls of the chamber contain narrow-band infra-red filters centred at 4.24 mu m (50% transmission points at 4.16 and 4.32 mu m) where CO2 absorption is high. A photoconductive infra-red sensor and an infra-red source are positioned on either side of the chamber. The output of the sensor is amplified by an instrumentation amplifier. Signal averaging of the sensor output in either the time or frequency domain was used to overcome the noise of the infra-red sensor. Step changes in Vc yielded exponentially changing outputs with a time constant of 1.1 ms. A quadrupole mass spectrometer's response to step changes in CO2 concentration generated in the measurement chamber fitted single exponential curves well with a maximum time constant of 37.7 ms and transport delay of 194 ms. A first-order model with delay fitted to the step response predicted the mass spectrometer frequency response well below 10 Hz but overestimated the response above 10 Hz. A third-order model with delay fitted to the frequency response predicted the step response very well. The author's results suggest that low-order models cannot predict the high-frequency performance of a mass spectrometer.