Water-column primary production was determined by the I4C in situ method during the spring bloom in the North Atlantic Ocean. For the same samples, the parameters of the photosynthesislight (P-I) curve were determined in broad-band light, and in narrow spectral bands for construction of the action spectrum. Using these parameters, with information on the vertical distribution of chlorophyll, measurements of light absorption by particulate materials, and data on surface irradiance, watercolumn production was calculated using 4 different production models. When compared to in situ primary production measurements, the results show that the spectral model, Model 1, is the best estimator of water-column primary production. Model 2 which used broad-band crB (the initial slope of P-I curve, normalized to biomass B) with light integrated over wavelength, and Model 4 (broad-band a* and broad-band light), consistently underestimated production by about 25 % and 60 % respectively. However, Model 3 (in which light is computed using a depth-averaged attenuation coefficient, R and in which a* is assumed to be wavelength-independent) gave water-column primary production estimates not significantly different from in situ values. It is recommended that the spectral model should be applied, whenever possible, in the computations of water-column primary production. If, however, broad-band crB has to be used in the calculations, it is suggested that light at depth be computed if possible using K The use of the fully broad-band model, Model 4, is not recommended. This is because the model gave strongly biased estimates of water-column primary production relative to the observed values.