Abstract. Global distributions of the CO₂ vmr (volume mixing ratio) in the mesosphere and lower thermosphere (from 70 up to  ∼  140 km) have been derived from high-resolution limb emission daytime MIPAS (Michelson Interferometer for Passive Atmospheric Sounding) spectra in the 4.3 µm region. This is the first time that the CO₂ vmr has been retrieved in the 120–140 km range. The data set spans from January 2005 to March 2012. The retrieval of CO₂ has been performed jointly with the elevation pointing of the line of sight (LOS) by using a non-local thermodynamic equilibrium (non-LTE) retrieval scheme. The non-LTE model incorporates the new vibrational–vibrational and vibrational–translational collisional rates recently derived from the MIPAS spectra by [Jurado-Navarro et al.(2015)]. It also takes advantage of simultaneous MIPAS measurements of other atmospheric parameters (retrieved in previous steps), such as the kinetic temperature (derived up to  ∼  100 km from the CO₂ 15 µm region of MIPAS spectra and from 100 up to 170 km from the NO 5.3 µm emission of the same MIPAS spectra) and the O₃ measurements (up to  ∼  100 km). The latter is very important for calculations of the non-LTE populations because it strongly constrains the O(³P) and O(¹D) concentrations below  ∼  100 km. The estimated precision of the retrieved CO₂ vmr profiles varies with altitude ranging from  ∼  1 % below 90 km to 5 % around 120 km and larger than 10 % above 130 km. There are some latitudinal and seasonal variations of the precision, which are mainly driven by the solar illumination conditions. The retrieved CO₂ profiles have a vertical resolution of about 5–7 km below 120 km and between 10 and 20 km at 120–140 km. We have shown that the inclusion of the LOS as joint fit parameter improves the retrieval of CO₂, allowing for a clear discrimination between the information on CO₂ concentration and the LOS and also leading to significantly smaller systematic errors. The retrieved CO₂ has an improved accuracy because of the new rate coefficients recently derived from MIPAS and the simultaneous MIPAS measurements of other key atmospheric parameters (retrieved in previous steps) needed for non-LTE modelling like kinetic temperature and O₃ concentration. The major systematic error source is the uncertainty of the pressure/temperature profiles, inducing errors at midlatitude conditions of up to 15 % above 100 km (20 % for polar summer) and of  ∼  5 % around 80 km. The errors due to uncertainties in the O(¹D) and O(³P) profiles are within 3–4 % in the 100–120 km region, and those due to uncertainties in the gain calibration and in the near-infrared solar flux are within  ∼  2 % at all altitudes. The retrieved CO₂ shows the major features expected and predicted by general circulation models. In particular, its abrupt decline above 80–90 km and the seasonal change of the latitudinal distribution, with higher CO₂ abundances in polar summer from 70 up to  ∼  95 km and lower CO₂ vmr in the polar winter. Above  ∼  95 km, CO₂ is more abundant in the polar winter than at the midlatitudes and polar summer regions, caused by the reversal of the mean circulation in that altitude region. Also, the solstice seasonal distribution, with a significant pole-to-pole CO₂ gradient, lasts about 2.5 months in each hemisphere, while the seasonal transition occurs quickly.