In this investigation we analyze two common optical configurations to retrieve CO2 total column amounts from solar absorption infrared spectra. The noise errors using either a KBr or a CaF2 beam splitter, a main component of a Fourier transform infrared spectrometer (FTIR), are quantified in order to assess the relative precisions of the measurements. The configuration using a CaF2 beam splitter, as deployed by the instruments which contribute to the Total Carbon Column Observing Network (TCCON), shows a slightly better precision. However, we show that the precisions in XCO2 ( = 0.2095 ⋅ Total Column CO2Total Column O2) retrieved from > 96 % of the spectra measured with a KBr beam splitter fall well below 0.2 %. A bias in XCO2 (KBr − CaF2) of +0.56 ± 0.25 ppm was found when using an independent data set as reference. This value, which corresponds to +0.14 ± 0.064 %, is slightly larger than the mean precisions obtained. A 3-year XCO2 time series from FTIR measurements at the high-altitude site of Altzomoni in central Mexico presents clear annual and diurnal cycles, and a trend of +2.2 ppm yr−1 could be determined.