Fourier transform infrared spectroscopy can be used for on-line smoke gas analysis of burning textile materials for several gases simultaneously. The technique is sensitive enough for the determination of the concentrations of compounds like H,O, CO,, CO, NO, NO,, SO,, C,H,O, HCI, HBr, HCN and HF. Quantitative analysis is, however, sometimes difficult to perform due to spectral interferences of two or more components. Simple mathematical techniques based on the direct relation between concentration and peak height or peak area then becomes impossible since they cannot be completely assigned to one component. Spectral subtraction can then be used but the results are sometimes unsatisfactory. Alternative techniques, like Partial Least Squares (PLS), provide excellent possibilities to overcome the problems due to overlapping spectral features, without making use of spectral subtraction. Robust calibration curves, based on PLS, can be constructed and used for prediction of unknown concentrations of different compounds in smoke gases. The use of PLS as a tool to overcome the problem of interfering components is demonstrated by application to the overlapping spectral bands of H,OICO, and of H,O/NO. PLS calibration curves for other interfering components can be constructed in an analogous way.