The National Centers for Environmental Prediction (NCEP) and National Center for Atmospheric Research (NCAR) have collaborated on a 50-year reanalysis project to produce a retrospective record of global atmospheric fields. This project involved recovering and quality controlling various data sources, including land surface, ship, rawinsonde, pibal, aircraft, satellite, and other observations, and assimilating them using a consistent data assimilation system over the reanalysis period. The reanalysis products include gridded fields, 8-day forecasts every 5 days, and a binary universal format representation (BUFR) archive of atmospheric observations. The reanalysis system includes a global spectral model with 28 vertical levels and a triangular truncation of 62 waves, equivalent to about 210-km horizontal resolution. The reanalysis data assimilation system is a three-dimensional variational (3DVAR) scheme cast in spectral space, denoted spectral statistical interpolation. The reanalysis is classified into three types: Type A variables are strongly influenced by observations and are the most reliable; Type B variables are influenced by both observations and the model and are less reliable; Type C variables are completely determined by the model and should be used with caution. The reanalysis has been affected by changes in the observing systems, particularly during the early period (1948-1957) and the introduction of satellite observations in 1979. The CD-ROM accompanying this article contains documentation and data on the reanalysis, including an observation data count program and a data file to assess observational coverage. The reanalysis has been used extensively in climate studies, and its impact on climatology and interannual variability is discussed. The paper also addresses known problems and errors in the reanalysis, such as human errors and model deficiencies, and highlights the importance of comparing reanalyses to assess reliability and trends. The NCEP-DOE Reanalysis 2 is a follow-on project aimed at correcting known problems and improving the reanalysis system.The National Centers for Environmental Prediction (NCEP) and National Center for Atmospheric Research (NCAR) have collaborated on a 50-year reanalysis project to produce a retrospective record of global atmospheric fields. This project involved recovering and quality controlling various data sources, including land surface, ship, rawinsonde, pibal, aircraft, satellite, and other observations, and assimilating them using a consistent data assimilation system over the reanalysis period. The reanalysis products include gridded fields, 8-day forecasts every 5 days, and a binary universal format representation (BUFR) archive of atmospheric observations. The reanalysis system includes a global spectral model with 28 vertical levels and a triangular truncation of 62 waves, equivalent to about 210-km horizontal resolution. The reanalysis data assimilation system is a three-dimensional variational (3DVAR) scheme cast in spectral space, denoted spectral statistical interpolation. The reanalysis is classified into three types: Type A variables are strongly influenced by observations and are the most reliable; Type B variables are influenced by both observations and the model and are less reliable; Type C variables are completely determined by the model and should be used with caution. The reanalysis has been affected by changes in the observing systems, particularly during the early period (1948-1957) and the introduction of satellite observations in 1979. The CD-ROM accompanying this article contains documentation and data on the reanalysis, including an observation data count program and a data file to assess observational coverage. The reanalysis has been used extensively in climate studies, and its impact on climatology and interannual variability is discussed. The paper also addresses known problems and errors in the reanalysis, such as human errors and model deficiencies, and highlights the importance of comparing reanalyses to assess reliability and trends. The NCEP-DOE Reanalysis 2 is a follow-on project aimed at correcting known problems and improving the reanalysis system.