We review results of calibrating daily probabilistic precipitation and temperature forecasts based on ensemble reforecast data sets for a newer (2005) ECMWF and older (1998) NCEP GFS models. We show that large training data sets are not necessary for the calibration of short-term temperature forecasts, but longer-lead temperature forecasts and precipitation forecasts benefit from large sample size. We also show that the decade-old GFS, after calibration, has more skill than the raw ECMWF forecasts. However, after calibration, the ECMWF forecasts are much more skillful than GFS calibrated forecasts, and a multi-model combination of the two is more skillful than either individually. We discuss the implications for operational usage of reforecasts.