Phase correction is a critical procedure for Fourier transform spectrometers(FTS), whose accuracy can be jeopardized from instrument properties and many uncontrollable environmental conditions, such as significant temperature change of interferometer, mechanical disturbances. Nevertheless, a generally applicable phase correction method seems not available, since current research result in limitations of the standard methods and propose solutions tailored to specific instruments. Considering of phase uncertainly and the challenge of phase determination, the phase property of FTS is characterization and analysis,then to resolve the FTS phase into an instrumental phase that is dependent on the interferometer temperature and a linear phase component that accounts for the discrepancy between the actual interferogram zero optical path difference (ZOPD) and the sampled one. The instrumental phase is mostly an instrumental characteristic that can be identified along with the other calibration parameters, the instrumental phase is strongly affected by interferometer temperature fluctuation, the instrumental phase is constant with interferometer temperature changes within threshold; While linear phase is main attributable to the offset between the digitized ZOPD and the real ZOPD, but also a mainly component of FTS phase. According to the principle of FTS, ZOPD sample errors can change abruptly from one interferogram to another; as a result, each spectrum has a different linear phase with respect to the wavenumber. So the phase processing can be simplification as a linear phase correction, residual phase, including the instrumental phase was removed in radiometric calibration later. This work considers the temperature properties of instrument phase, the implemented method is based on the identification of linear phase by least-squares approach, with interferogram symmetrization, residuals phase of measured data is stabilization to permit spectra averaging, so the artifacts due to vibrations are removed, the last step is complex radiometric calibration procedure with mostly instrumental phase removed, and then the phase correction is accomplished. Based on experimental data, the linear phase terms are derived by the leasts-quares method and removed. Data were interpolated to model the instrumental phase using different order polynomials. A fifth-order polynomial fitting was eventually used with mean square errors(MSE) of 0.13 rad, because of a further reduction of MSE is not entail with the order increased. Instrumental phase with interferometer temperatures of 283, 290 and 300 K were taken by fifth-order polynomial fitting. Phase regression residuals with the instrumental phase and linear phase removed, stochastic distributed central at zero, the temperature dependence of the instrumental phase by theoretic is demonstrated by experiment. And then, the feasibility of the proposal phase processing method is experimentally validated. Experimental result shows that phase residuals within ±0.04 rad and the radiometric uncertainty sought 0.8 K accuracy or better is achieved.