We have developed a procedure for retrieving atmospheric abundances of HFC-23 (CHF3) with a ground-based Fourier transform infrared (FTIR) spectrometer and analyzed the spectra observed at Rikubetsu, Japan (43.5 °;N, 143.8 °;E), and at Syowa Station, Antarctica (69.0 °;S, 39.6 °;E). The FTIR retrievals were carried out with the SFIT4 retrieval program, and the two spectral windows of 1138.5-1148.0 cm-1 and 1154.0-1160.0 cm-1 in the overlapping ν2 and ν5 vibrational-rotational transition bands of HFC-23 were used to avoid strong H2O absorption features. We considered O3, N2O, CH4, H2O, HDO, CFC-12 (CCl2F2), HCFC-22 (CHClF2), peroxyacetyl nitrate (PAN) (CH3C(O)OONO2), HCFC-141b (CH3CCl2F), and HCFC-142b (CH3CClF2) to be interfering species. Vertical profiles of H2O, HDO, and CH4 are preliminarily retrieved with other independent spectral windows because these profiles may induce large uncertainties in the HFC-23 retrieval. Each HFC-23 retrieval has only one piece of vertical information with sensitivity to HFC-23 in the troposphere and the lower stratosphere. Retrieval errors mainly arise from the systematic uncertainties of the spectroscopic parameters used to obtain HFC-23, H2O, HDO, and CH4 abundances. For comparison between FTIR-retrieved HFC-23 total columns and surface dry-air mole fractions provided by AGAGE (Advanced Global Atmospheric Gases Experiment), FTIR-retrieved HFC-23 dry-air column-averaged mole fractions (XHFC-23) were calculated. The FTIR-retrieved XHFC-23 values at Rikubetsu and Syowa Station have negative biases of -15 % to -20 % and -25 % compared to the AGAGE datasets, respectively. These negative biases might mainly come from systematic uncertainties of HFC-23 spectroscopic parameters. The trend of the FTIR-retrieved XHFC-23 data at Rikubetsu was derived for December to February (DJF) observations, which are considered to represent the background values when an air mass reaching Rikubetsu has the least influence by transport of HFC-23 emissions from nearby countries. The DJF trend of Rikubetsu over the 1997-2009 period is 0.810 ± 0.093 ppt yr-1 (ppt: parts per trillion), which is in good agreement with the trend derived from the annual global mean datasets of the AGAGE 12-box model for the same period (0.820 ± 0.013 pptyr-1). The DJF trend of Rikubetsu over the 2008-2019 period is 0.928 ± 0.108 pptyr-1, which is consistent with the trend in the AGAGE in situ measurements at Trinidad Head (41.1 °;N, 124.2 °;W) for the same period (0.994 ± 0.001 pptyr-1). The trend of the FTIR-retrieved XHFC-23 data at Syowa Station over the 2007-2016 period is 0.819 ± 0.071 pptyr-1, which is consistent with that derived from the AGAGE in situ measurements at Cape Grim (40.7 °;S, 144.7 °;E) for the same period (0.874 ± 0.002 pptyr-1). Although there are systematic biases in the FTIR-retrieved XHFC-23 at both sites, these results indicate that ground-based FTIR observations have the capability to monitor the long-term trend of atmospheric HFC-23. If this FTIR measurement technique were extended to other Network for the Detection of Atmospheric Composition Change (NDACC) ground-based FTIR sites around world, the measurements reported from these sites would complement the global AGAGE observations by filling spatial and temporal gaps and may lead to improved insights about changes in regional and global emissions of HFC-23 and its role in global warming.