This paper focuses on the accuracy of longwave radiation flux retrievals at the top and bottom of the atmosphere at Eureka station, Canada, in the high Arctic. We report comparisons between seven products derived from (1) calculations based on a combination of ground-based and space-based lidar and radar observations, (2) standard radiometric observations from the CERES satellite, (3) direct observations at the surface from a broadband radiation station and (4) the ERA-Interim and ERA5 reanalyses. Statistical, independent analyses are first performed to look at recurring bias and trends in fluxes at Top and Bottom of the Atmosphere. The analysis is further refined comparing fluxes derived from coincident observations decomposed by scene types. Results show that radiative transfer calculations using ground-based lidar-radar profiles derived at Eureka agree well with TOA LW fluxes observed by CERES and with BOA LW fluxes reference. CloudSat-CALIPSO also show good agreement with calculations from ground-based sensor observations, with a relatively small bias. This bias is shown to be largely due to low and thick cloud occurrences that the satellites are insensitive to owing to attenuation from clouds above and surface clutter. These conditions of opaque low clouds, cause an even more pronounced bias for CERES BOA flux calculation in winter, due to the deficit of low clouds identified by MODIS. ERA-I and ERA5 fluxes behave differently, the large positive bias observed with ERA-I is much reduced in ERA5. ERA5 is closer to reference observations due to a better behaviour of low and mid-level clouds.