The statistical characteristics of the systematic and random observation errors of the line-of-sight (LOS) velocity measurements made by the high-resolution Doppler imager (HRDI) are obtained using collocated and coincident differences between HRDI observations and model-assimilated data (H-A). Although these differences have contributions from both observation and model errors, we show how they can be individually characterized. HRDI, a limb sounder on board the Upper Atmosphere Research Satellite (UARS), is the only instrument making global wind measurements in the stratosphere. Doppler shifts in O2 absorption spectra measured by HRDI within two different bands (γ and B) basically provide the line-of-sight velocity. The analysis presented here is carried out using the Doppler shifts themselves, rather than the HRDI retrieved (u, v) wind components. The absolute position of the line for zero-wind conditions in LOS velocity measurement is unknown, and one goal of this paper is to develop a systematic method of determining this position through mean differences (biases) between observed HRDI and LOS model analyses approximately projected onto the line of sight. Biases are separated based on 16 different measurement configurations. The bias dependence on these configurations (plus an altitude dependence) is used to create a bias correction for each instrument configuration. The random error component in H-A differences is analyzed by separating the γ and B band observations and locating observation pairs where both bands are very nearly looking at the same location at the same time. It is shown that the two observation error streams are uncorrelated. Observation errors are also uncorrelated with the analyses error, and observational and model wind variance (in the stratosphere) are estimated.