The dataset contains physical, biological and chemical oceanographic measurements, and meteorological data. Hydrographic measurements include temperature, salinity, current velocities, attenuance, dissolved oxygen and fluorescence, while water samples were analysed for concentrations of nutrients, pigments, suspended particulates, metals and halocarbons. Samples were also collected for phytoplankton and zooplankton analyses, while results from production experiments are also included in the data set. These oceanographic data are supplemented by surface meteorological measurements. The data were collected at 357 sites in the NE Atlantic, 308 of which are from cruises centering on 20 W, 47 to 60 N, 16 from the Cape Verde Islands and 33 in a coccolithophore bloom just south of Iceland. Measurements were taken from 3 cruises in 1989, 6 cruises in 1990 and 2 cruises in 1991. The data were collected via (i) underway sampling (SeaSoar Undulating Oceanographic Recorder (UOR), hull-mounted acoustic Doppler current profiler (ADCP), meteorology and surface ocean parameters) of which there are 793430 records at 30 second intervals from 11 cruises and (ii) discrete sampling (conductivity-temperature-depth (CTD) and expendable bathythermograph (XBT) casts, bottle stations, net hauls, productivity incubations, stand alone pump (SAP) and sediment trap deployments, cores) of which there are 2215 deployments/experiments. The aim of the Biogeochemical Ocean Flux Study (BOFS) Community Research Project was to study the role of oceans in the global cycling of carbon. The data were collected and supplied by UK participants in the Joint Global Ocean Flux Study (JGOFS). The British Oceanographic Data Centre (BODC) had responsibility for calibrating, processing, quality controlling and documenting the data and assembling the final data set. The underway data are stored as time series for each cruise merged with the navigation data. The data are fully quality controlled. Checks were made for instrument malfunction, fouling, constant values, spikes, spurious values, calibration errors and baseline corrections. The discrete data are stored in a relational database (Oracle RDBMS), mainly as vertical profiles and are uniquely identified by a combination of deployment number and depth.