Ground-based microwave radiometry is a common tool to estimate profiles of the atmosphere. With a high temporal resolution radiometers have became an alternative to atmospheric sounding like radiosondes. However remote sensing radiometry requires the use of inversion algorithms, where methods like linear-, quadratic-regression or Artificial Neural Network are commonly used. The present study implements a Bayesian inversion technique as alternative to the state-of-the-art retrieval algorithms provided by the radiometer’s manufacturer firmware. The Bayesian inversion provides advantages over other established methods, namely: the use of a-priori suited for the specific climatology under observation, the estimation of the most likely profile along with its uncertainty obtained from the posteriori distribution, and the feasibility to add synergistic observations from other instruments to increase retrieval capabilities. To estimate the uncertainties resulting from the Bayesian and firmware retrieval algorithms, synthetic radiometer data have been created by means of radiative transfer simulations using radiosonde profiles as descriptor of atmospheric states. These synthetic data mimics the instrument’s firmware binary files letting the radiometer to perform retrievals as real measurements. By analyzing the differences from retrieval results relative to the known true profile we assess uncertainty metrics to characterize the algorithms. It has been found that Bayesian inversion reproduces more accurately the profile vertical structure as compared to the firmware, specially for humidity profiles. Absolute errors have been strongly reduced mainly at the lower atmosphere. The study concludes that Bayesian inversion for ground-based atmospheric profiling produces results resembling observations by radiosondes when a suitable a-priori distribution is used.