A statistical mode! is developed to reconstruct atmospheric surface data for the period 1902-I998 to force a coupled ocean-sea ice mode! of the Baltic Sea. As the longest time scale of the Baltic Sea is on the order of 30 years, climate relevant model studies should cover century Iong simulations. One of the aims of the Swedish Regional Climate Modeling Program is to investigate causes of longterm changes of the Baltic Sea during the past 100 years. However, an observational atmospheric dataset for such a purpose is not available yet. High resolution in space and time is required. Therefore, we developed a statistical mode! using a Redundancy Analysis to reconstruct daily sea leve! pressure and monthly surface air and dew-point temperature, precipitation, and cloud cover fields on a 1o x I o regular horizontal grid for the Baltic Sea region. A gridded atrnospheric data set based on synoptic stations, which is available for the period 1970-2001, is used as input to the statistical mode!. Spatial patterns are selected by maximizing predictand variance during the "learning" period 1980-1998. The remainder period 1970-1979 is used for validation. It is shown that the statistical model works reasonable well for all variables. We found the highest skill for sea leve! pressure and the lowest skill for cloud cover.