A technique of analysis for RS-FTIR (Remote Sensing Fourier Transform Infrared) spectrum was built based on the principle of calibration transfer. A system of four-component gas mixtures, whose FTIR spectra were overlapped seriously, was determined as the object in the present research. Its FTIR data from EPA and field measurement were used to set up calibration and prediction models, respectively. With OSC (orthogonal signal correction), the model built with data of EPA predicted values of field measurement. The RMSEP values of acetone, methanol, benzene and chloroform were 0.0085, 0.0180, 0.0640 and 0.0028, respectively, while without OSC correction the corresponding values were 0.0856, 0.0479, 1.0653 and 0.0142, respectively. With optimization, the Kennard-Stone method was used for subset selection, and the iteration in the process of OSC was carried out 3 times, yielding the best performance. The result indicates that this technique can overcome the limitation of background and calibration model, which restricts the application of RS-FTIR in air monitoring.