Global cloud parameters including cloud liquid water, cloud base height, cloud top height and cloud type are observed from the Cloud Profiling Radar (CPR) on board CloudSat. Gobal COSMIC Radio Occultation (RO) data during a three year period from 2007-2009 are collocated in space and time with CloudSat data. The collocated dataset is then classified into eight groups: one clear-sky condition and seven different cloud types. For each group, atmospheric refractivity, temperature and water vapor derived from COSMIC GPS ROs are compared with those of the European Centre for Medium-Range Weather Forecasts (ECMWF) analyses. It is found that the analysis fields of the refractivity (N) have a systematic positive bias under all cloudy conditions. The fractional N-bias is as high as 1.2% depending on cloud types. Using CloudSat liquid water content (LWC), it is demonstrated that LWC can contribute 0.8% of the total refractivity by individual clouds. The presence of liquid water content could contribute to the positive N-bias with a magnitude of 0.16%. This value is slightly greater than the retrieval uncertainty quantified by the mean difference between the observed refractivity with LWC contribution subtracted and the refractivity calculated by GPS retrievals. The GPS RO retrieval uncertainty increases linearly with LWC as anticipated theoretically. Using all COSMIC ROs and CloudSat observations within the three years, it is demonstrated that positive N-biases are spatially correlated with positive water vapor biases, negative temperature biases, and large liquid water contents.