Monthly mean fields of temperature and geopotential height (GPH) from 700 to 100 hPa were statistically reconstructed for the extratropical Northern Hemisphere for the World War II period. The reconstruction was based on several hundred predictor variables, comprising temperature series from meteorological stations and gridded sea level pressure data (1939-1947) as well as a large amount of historical upper-air data (1939-1944). Statistical models were fitted in a calibration period (1948-1994) using the NCEP/NCAR Reanalysis data set as predictand. The procedure consists of a weighting scheme, principal component analyses on both the predictor variables and the predictand fields and multiple regression models relating the two sets of principal component time series to each other. According to validation experiments, the reconstruction skill in the 1939-1944 period is excellent for GPH at all levels and good for temperature up to 500 hPa, but somewhat worse for 300 hPa temperature and clearly worse for 100 hPa temperature. Regionally, high predictive skill is found over the midlatitudes of Europe and North America, but a lower quality over Asia, the subtropics, and the Arctic. Moreover, the quality is considerably better in winter than in summer. In the 1945-1947 period, reconstructions are useful up to 300 hPa for GPH and, in winter, up to 500 hPa for temperature. The reconstructed fields are presented for selected months and analysed from a dynamical perspective. It is demonstrated that the reconstructions provide a useful tool for the analysis of large-scale circulation features as well as stratosphere-troposphere coupling in the late 1930s and early 1940s.