Model based observation of carbon dioxide (CO₂) in a burning process is discussed. The described model structure is a combination of fuzzy Takagi-Sugeno (TS) models and operation regime based modelling approach. The selection of local modelling regions and input variables is based on general combustion theory. Recursive stochastic gradient method is employed to model training. Simulations with experimental data are analysed to verify the validity of the discussed combustion observation approach. Performance is further compared to neural and linear models. Results indicate that the presented model has good generalisation properties and it is capable to capture systems behaviour.