We are addressing this trillion-dollar issue as cleanly as is currently feasible using large ensembles of three distinct types of observational and model simulation datasets for the 1874 to 2010 period. The first type is a 56-member ensemble of global atmospheric reanalyses at 6-hourly resolution, the 20CRv2 dataset. These reanalyses are based on assimilating surface pressure observations in an ensemble Kalman filter based assimilation system, in which short-term forecasts using an atmospheric GCM with specified observed SSTs and sea ice and radiative forcings are provided as background first guess fields at each analysis time. The second type of dataset, that we refer to as the AMIP20C dataset, is a 56-member ensemble of atmospheric GCM simulations of the 1874 to 2010 period generated by us using the same atmospheric GCM used to produce the 20CRv2 dataset, and with identical specifications of time-varying SST, sea ice, and radiative forcings. The third dataset type is a multi-model ensemble of 62 CMIP5 coupled climate model simulations of the same 1874 to 2010 period with observed radiative forcings. These three types of datasets at daily resolution for the identical 1874 to 2010 period enable us to make cleaner separations of radiatively forced versus internal natural climate variability than has previously been possible. We do this by interpreting the long-term variability in the 20CRv2 observational dataset as a combination of internal chaotic, SST-forced, and radiatively forced variations; the variability of the ensemble-mean AMIP20C responses as a combination of SST-forced and radiatively forced variations; and the variability of the CMIP5 multi-model ensemble-mean responses as radiatively forced variations. An important new result is that the observed strong trends of many circulation variables over the second half (1943-2010) of the record are much weaker or non-existent when calculated over the full record (1874-2010). This is especially true for northern hemispheric trends. Consistent with the weak long-term circulation trends, we find that the long-term storminess trends in the AMIP20C and CMIP5 simulations are also weak. This weakness of the circulation and storminess trends over the full 136-yr record has important implications for our understanding of the atmospheric circulation response to global warming, and casts doubt on inferences about this response drawn in numerous published studies from considering only the second half of the record.