We have stacked planktonic carbon isotope data from three cores in the western equatorial Pacific in order to generate a new reconstruction of atmospheric carbon dioxide over the past 450,000 years. Our new reconstruction resembles that of Shackleton et al. (1983) based on data from East Pacific core V19-30, which successfully predicted features that were subsequently verified by Barnola et al. (1987) in the record from the Vostock ice core. In addition the new data confirm the discovery of Shackleton and Pisias (1985) that changes in atmospheric CO 2 lead changes in ice volume and hence probably contributed to the glacial-interglacial cycles. Our new reconstruction avoids some of the deficiencies of the previous reconstruction: in particular the planktonic species ( Neogloboquadrina dutertrei), on which the earlier reconstruction depends, does not calcify in truly nutrient-free surface water as the model assumes, whereas our new reconstruction uses Globigerinoides sacculifer which is expected to be more reliable. In addition, the surface waters in the west Pacific are closer to the nutrient-free ideal on which the model (Broecker, 1982) depends. On the other hand, the amplitude of the new reconstruction is significantly smaller than the amplitude observed by Barnola et al. (1987). It is not clear whether this smaller range is a better estimate of the amplitude of the ‘biological pump’ effect, or whether the true amplitude is reduced by bioturbation in the west Pacific cores that we studied.