An extensive analysis of radiative flux biases in the Climate Forecast System Model Version 2 (CFSv2) is done. Annual mean and seasonal variations of biases at the surface and top of the atmosphere (TOA) are reported in the global domain. Large regional biases in shortwave (SW) and longwave (LW) radiation are observed over convectively active zones in the tropics. The relative contribution of various processes responsible for the reported biases is quantified. The poor simulation of clouds and inadequate representation of surface properties seem to be major contributors. Over certain regions, errors due to different processes add up, whereas, over other regions, errors tend to nullify each other. Surface and atmospheric variables taken as input parameters in the radiative transfer modules are compared with satellite-based observations. The maximum biases in SW and LW radiation are observed over the regions of persistent low clouds. The magnitude of the SW and LW biases at the TOA is in phase with the biases in cloud fraction by and large. However, the error in the radiative fluxes due to errors in surface radiative properties is of equal importance. The cold bias in near-surface air temperature reported in other studies may partly be attributed to an underestimation in the net SW radiation at the surface. In the present study, a plausible prescription is also provided to correct the source of the biases. © 2020, Springer-Verlag GmbH Germany, part of Springer Nature.