We investigated a variety of factors that influence the determination of precipitable water vapor (V) and integrated cloud liquid (L) by dual-channel microwave radiometers (MWRs). These factors include radiometric calibration; dry, water vapor, and liquid absorption coefficients; and physical versus statistical retrieval methods. We then applied the analysis to the MWR that was operated by the Atmospheric Radiation Measurement Program (ARM) during the Surface Heat Budget of the Arctic Ocean project. In the work reported here, MWR data taken from April 1 to July 31, 1998, were analyzed. Data acquired in situ did not always agree with the original MWR liquid retrievals, with MWR estimates at times being too large by perhaps a factor of 2. These differences led us to examine in detail several of the assumptions that go into V and L retrievals. The radiometer was carefully examined and found to be well calibrated with a 0.3 K RMS error. The predicted accuracy in the L retrievals, for this 0.3 K RMS radiometric error and a statistical retrieval, was 25 g m-2 RMS. This accuracy improves to 10%, if we use the improved knowledge of cloud temperature, as can be obtained using radiosondes and cloud radar/lidar measurements. We also studied the degree to which different clear air and cloud liquid models have an effect on V and L retrievals. The most significant changes from the original ARM retrievals were due to the dry opacity and the cloud liquid dielectric model. Although nothing was found in the original ARM data that was grossly incorrect, application of these models reduced the original ARM retrievals by roughly 20 to 30%. The change of clear-air absorption model from the original to a more recent one has little impact on V retrievals except when V is <0.5 cm. ¿ 2001 American Geophysical Union