We make use of a simple diagnostic box model to determine the sensitivities of the influencing parameters for the isotopic signal transfer of seawater oxygen to atmospheric oxygen. We calculate the Δ18O of atmospheric oxygen from prescribed oxygen fluxes of the living and dead biomes on land and in the ocean, respectively. The model is driven by an assumed (experiment 1) or measured (experiments 2 and 3) temporal seawater Δ18O signal and a land biomass estimation. In experiment 1, we calculated the required changes of several model parameters in order to study fast variations of Δ18O of atmospheric oxygen as seen in the Greenland Ice Core Project (GRIP) ice at depths assigned to the Eemian time period. Our calculations support evidence of stratigraphic problems at these depths in the GRIP ice core. In experiment 2, we adjusted the model output, which was driven by the benthic seawater Δ18O record from V19--30, to the measured Greenland Ice Sheet Project 2 Δ18O record of atmospheric oxygen for the last 110,000 years, by varying the model parameter. Single and multiparameter matchings were performed. The results for single-parameter runs exceed the uncertainty ranges for most of the parameters, while multiparameter variations are well within these ranges. The model calculations are most sensitive to the land respiration factor. Our results support the findings of Van de Water et al. <1994> that the fractionations associated with biomes activities were most probably lower during cold periods, which could point to a combination of fractionations with different temperature dependencies. The model results indicate periods of higher marine biological activity during the ice age than today. Temporal variations of the model parameters show a double peak around 10000 and 8000 years ago, which could be associated with meltwater pulses, as shown in experiment 3. However, they are hardly the well-known Fairbanks <Fairbanks et al., 1992> pulses since these occur 3000 to 4000 years earlier. ¿ 1997 American Geophysical Union