A whole body indirect calorimeter provides accurate measurement of energy expenditure and the respiration quotient over long periods of time, but has limitations to assess dynamic changes in energy metabolism due to the small amplitude of the signal in relation to the size of the room. The present study aimed to improve algorithm for the whole body indirect calorimetry by adopting a deconvolution with a regularization parameter. Performance of the new algorithm was compared with trends identification (Med. & Biol. Eng. & Comput 34 : 212, 1996) against four validation tests. In a simulated problem, in which metabolic rate cycled with a period of 20 min, mean square errors (MSEs) computed at every 1 min by the deconvolution (0.0036 for O₂ consumption and 0.0017 for CO₂ production) were smaller than those for trends identification algorithms (0.0198 and 0.0142). Deconvolution algorithm clearly separated individual CO₂ infusion of 8 min intervals, while trends identification could no longer separate them. During the validation by ethanol combustion, which produced a near-steady state condition, the deconvolution (MSEs were 0.0022 for O₂ consumption and 0.0010 for CO₂ production) performed better than trends identification algorithms (MSEs were 0.0086 and 0.0041). When validated against direct measurement of gas production rate during non-steady state condition, produced by a human subject intermittently exercising in the calorimeter, deconvolution (MSEs were 0.0032 for O₂ consumption and 0.0038 for CO₂ production) performed better than trends identification algorithms (0.0182 and 0.0167). The new algorithm significantly improved transient response of the whole body indirect calorimeter.