CO$_2$ mixing ratio derived from spaceborne measurements of the Television Infrared Observation Satellite (TIROS-N) Operational Vertical Sounder (TOVS) instrument onboard NOAA-10 available for the time period 1987 - 1991 are evaluated against modeling results and aircraft measurements. The model simulations are based on two transport models and two sets of surface fluxes which have been optimized in order to fit near-surface atmospheric CO$_2$ measurements through a transport model ( using an inverse procedure). In the tropics the zonal mean annual cycle and growth rate of the satellite product are consistent with those of the models. However, north-to-south gradients and spatial distributions for a given month show large differences. There are large regional patterns that can reach 7 ppm in the satellite retrievals ( over regions of a few thousand kilometers wide) but are absent in the model predictions. The root-mean-square (RMS) differences between the models and the satellite product are around 1.7 ppm. One time series of the model CO2 trend is used to extrapolate to the airborne measurement periods ( 1991 - 2003) both the satellite and the model monthly products to the airborne measurement period. The RMS difference between the airborne measurements and the extrapolated model predictions is around 1 ppm, while it is 2 ppm for the satellite estimate. These comparisons suggest that the large spatial variability of TOVS retrievals reflects substantial regional biases and noise which need to be reduced before remotely sensed CO$_2$ from TOVS will help constrain our knowledge of the carbon cycle.