The long-term variability in the terrestrial and oceanic uptake of anthropogenic carbon is investigated. Ice core and direct observations of atmospheric CO2 and 13C are used for the last 200 years. An inverse method called double deconvolution is applied. It is found that the biosphere turned from a carbon source of about 0.5 Gt C yr-1 into a sink of 1 Gt C yr-1 during the first half of this century. This is in qualitative agreement with earlier reconstructions based on atmospheric CO2 data and implies a terrestrial sink to compensate land use emissions during the last five decades. Oceanic and biospheric carbon uptakes are estimated to be 0.9¿1.0 and 1.1¿1.0 Gt C yr-1 as averaged over the 1970--1990 period. Hence ocean uptake is on the low side of current estimates, but our results may be biased as &dgr; 13C observations between 1956 and 1982 are missing. Additional uncertainties in calculated carbon sinks are due to uncertainties in model parameters and in fossil emission estimates. Prior to 1950, uncertainties are primarily related to uncertainties in the ice core &dgr; 13C data; a Monte Carlo analysis yields a 1-&sgr; uncertainty in the terrestrial and oceanic uptake of ¿0.36 Gt C yr-1 when ice core data are smoothed over a 50 year period. The budget of bomb-produced radiocarbon is reinvestigated. We could not find model solutions that concomitantly match the bomb budget and the observed atmospheric &dgr; 13C and prebomb &Dgr; 14C decrease. The closure of the budget would require a relatively low oceanic and biospheric 14C uptake that conflicts with the relatively high isotopic uptake rates required to simulate the observed decrease in &dgr; 13C and &Dgr; 14C. We conclude that recent estimates of bomb test productions and/or the stratospheric 14C decrease are not compatible with published 13C and 14C observations. ¿ 1998 American Geophysical Union