It has been noted that the ERBE cloud radiative forcing (CRF) has a significant bias of up to about 20 Wm-2 over the tropical moist region, compared model analysis results (e.g., Collins and Inamdar, 1995; Slingo et al., 1998). However, it has never been clarified why there are such large differences and where those systematic differences are due to. Here we provide evidences that those discrepancies are caused by different definitions of cloud forcing (or clear-sky fluxes) employed by two different approaches. Model analysis employs CRF defined as the radiation changes by clouds during the transition from cloud-free state to cloudy state of the earth-atmosphere system, while other surface atmospheric variables are held fixed during the transition. In contrast, ERBE CRF implicitly includes additional radiative effects due to the variations of surface and atmospheric variables that are correlated with clouds. A positive relationship between SSM/T-2 derived upper tropospheric water-vapor (UTW) and collocated CERES CRF strongly supports the notion that the aforementioned biases are largely due to different distributions of upper tropospheric water vapor employed in the model analysis and ERBE clear-sky climatologies. We computed a possible range of the CRF bias using a relationship between ISCCP high cloud and SSM/T-2 derived UTW. Up to 20 Wm-2 of CRF bias appears to be contributed by UTW change alone over convectively active regions, indicating that at most one quarter of ERBE CRF is the water vapor forcing associated with high cloud development.