We have updated earlier deconvolution analyses using most recent high-precision ice core data for the last millennium (Etheridge et al., 1996) and direct atmospheric CO2 observations starting in 1958 (Keeling and Whorf, 1994). We interpreted nonfossil emissions, that is, the difference between the increase in observed atmospheric plus modeled oceanic carbon inventory and fossil emissions, as biospheric carbon storage (release). We have assessed uncertainties in the CO2 ice core data using a Monte Carlo approach and found a 2-&sgr; uncertainly for the nonfossil emissions (20-year averages) of 0.2--0.4 Gt C yr-1. Overall uncertainties of the nonfossil emissions were estimated to be 0.5 Gt C  yr-1 before 1950 and ~1 Gt C  yr-1 during the last decade. We found a large and rapid change of -0.8 Gt C yr-1 in the nonfossil emissions (approximately net air-biota flux) between 1933 and 1943. Before 1933, the land biota acted as carbon source of order 0.5 GtC yr-1 in agreement with independent estimates of carbon emissions by land use changes (Houghton, 1993a). After 1943, the land biota was a net sink of about 0.3 GtC  yr-1. This implies an average biospheric sink of 1.5 GtC yr-1 during the last 5 decades to compensate estimated carbon emissions by land use changes. We could not attribute this sink to a single mechanism. We found that the temporal evolution of the required biota sink is not compatible with conventional modeling of CO2 fertilization. We estimated potential terrestrial carbon storage due to nitrogen fertilization to be 1 GtC yr-1 for 1960, that is, smaller than the required sink, and 1.5--3 GtC yr-1 for 1990. To assess the potential impact of climate variations, we deconvolved the preindustrial CO2 concentrations which fluctuated around 280 ppm. We found a maximum nonfossil sink of 30 GtC within 50 years. Thus it seems not likely that the cumulative sink of 76 Gt C which is required to balance land use emissions during 1935 to 1990 can be explained by climate variations only.¿ 1997 American Geophysical Union