Advances in macroscale land surface modeling, and rapidly improving computational capabilities, facilitate the implementation of physically realistic land surface parameterizations that represent water and energy fluxes over large areas. Evaluation of these formulations has been undertaken by recent efforts such as the GEWEX Project for Intercomparison of Land Surface Parameterizations (PILPS), and Water and Energy Budget Synthesis study (WEBS). In all such evaluations, a baseline of observations, or benchmark simulations, is needed with which simulated land surface states and fluxes can be compared. We describe a retrospective North American-Land Data Assimilation System (N-LDAS) land surface data set, produced at 1/8 degree spatial resolution and 3-hourly time resolution over the N-LDAS domain (North America between latitudes 25 to 53 north) for the period January 1950 through July 2000. The data set consists of both surface hydrologic forcings, and surface energy and moisture fluxes and state variabiles. The former are derived more or less directly from observations (downward solar and longwave radiation, precipitation, temperature, relative or absolute humidity, surface wind, and surface pressure). Moisture and energy fluxes (runoff, evapotranspiration, reflected solar and emitted longwave radiation, sensible and ground heat flux) and state variables (surface temperature, soil moisture, and snow water storage) are derived using the Variable Infiltration Capacity (VIC) macroscale hydrology model. Gridded precipitation and temperature are derived from NOAA Cooperative Observer daily station data (over 12,000 stations) disaggregated to three-hourly values, while surface humidity and radiative forcings are derived from daily temperature and temperature range. Surface wind is taken from NCEP/NCAR reanalysis. The VIC model output is included with the forcing data which constitute a GCIP legacy data set available from UCAR in one of three forms: monthly summary, daily summary, and the full three-hourly data set.