Several recent studies have observed biases between measurements in the 183.31GHz water vapour line by space-borne sounders and calculations using radiative transfer models with inputs from either radiosondes (RAOBS) or short range forecasts by Numerical Weather Prediction (NWP) models. This paper discusses all the relevant categories of observation-based or model-based data, quantification of their uncertainties, and separation of biases that could be common to all causes from those attributable to a particular cause. Reference observations from radiosondes, Global Navigation Satellite Systems (GNSS), Differential Absorption Lidar (DIAL) and Raman lidar are thus overviewed. Biases arising from their procedures of calibration, NWP models and data assimilation, instrument biases and radiative transfer models (both the models themselves and the underlying spectroscopy) are presented and discussed. Although no single process in the comparisons seems capable of explaining the observed structure of bias, recommendations are made in order to better understand the causes.