This work presents statistically reconstructed global monthly mean fields of temperature and geopotential height (GPH) up to 100 hPa for the period 1880–1957. For the statistical model several thousand predictors were used, comprising a large amount of historical upper-air data as well as data from the earth’s surface. In the calibration period (1958–2001), the statistical models were fit using the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) as the predictand. After the weighting of the predictors, principal component (PC) analyses were performed on both the predictand and predictor dataset. Multiple linear regression models relate each principal component time series from the predictand with an optimal subset of principal component time series from the predictor. To assess the quality of the reconstructions, statistical split-sample validation (SSV) experiments were performed within the calibration period. Furthermore, the reconstructions were compared with independent historical upper-air and total ozone data. Based on the SSV experiment, this study obtained good reconstructions for temperature and GPH in the Northern Hemisphere; however, the skill in the tropics and the Southern Hemisphere was much lower. The reconstruction skill shows a clear annual cycle with the highest values in January. The lower levels were better reconstructed except in the tropics where the highest levels showed the best skill. With the inclusion of a considerable amount of upper-air data after 1939 the skill increased substantially. The fields were analyzed for selected months in the 1920s and 1930s to demonstrate the usefulness of the reconstructions. It is shown that the reconstructions are able to capture regional-to-global dynamical features.