Climate research relies on realistic atmospheric data over long periods of time. Global reanalyses or observations are commonly used for this type of work. However, the many problems associated with both the reanalyses and observations cast doubts on the reliability of such data for climate applications, and users often need to know how large the errors and uncertainties associated with the different data sets are. This paper is a systematic assessment of the errors and uncertainties contained in the time mean (1979–1999) of many different climate quantities taken from a variety of global data sets, including four popular reanalyses, the output of the climate model developed at the Geophysical Fluid Dynamics Laboratory (GFDL), and a wide range of observations. We find that the ability of reanalyses to reproduce the observed climate mean state varies widely, with radiative quantities exhibiting the largest discrepancies. The different reanalysis products share many common errors, but overall the European Centre for Medium-Range Weather Forecasts 40-year reanalysis (ERA-40) matches best the observations. Interestingly, the climate model reproduces the observed climate mean state of certain quantities more faithfully than the reanalyses. This indicates that modern models have reached a high level of realism in their mean state and that care must be taken when reanalyses are used to validate models. A particular concern of this paper is the time mean uncertainty associated with specific observation-based atmospheric quantities. Observational uncertainties are estimated from the difference amongst alternative data sets for the same quantity. We show that for most quantities the observational uncertainty is smaller than the error of the reanalyses or the model. However, there are some notable exceptions. In particular, for the surface fluxes of heat, momentum, and radiation the observational uncertainties can be as large as the errors seen in the reanalyses or the model. The investigation of uncertainties in upper atmospheric quantities is restricted to reanalysis and model data, since no appropriate observations are available. In this case, the reanalyses uncertainties are generally smaller than the model errors, except for quantities which describe the meridional component of the atmospheric circulation.