Abstract: The main objective of this thesis is to understand the millennial variability of atmospheric CO2 during the Marine Isotope Stage 6 (MIS 6), the penultimate glacial, period (185-135 kyr BP). During the early MIS 6 period (185-160 kyr BP), 6 millennial-scale climate oscillations can be observed in proxy records of Antarctic temperature, the bipolar see-saw phenomenon in the North Atlantic region, and Monsoon intensity in low latitudes. An intensified hydrological cycle and iceberg calving in the North Atlantic may have impacted on the Atlantic Meridional Overturning Circulation during MIS 6 (Margari et al., 2010). Atmospheric CO2 reconstructions from Antarctic ice cores can provide key information on how atmospheric CO2 concentrations are linked to millennial-scale climate changes. However, existing CO2 records from the Vostok ice core do not show the millennial variability due to the lack of suitable temporal resolution and precision. To understand atmospheric CO2 variability during MIS 6, a precision of less than 2 ppm is mandatory, because there is a possibility that we could observe small CO2 variability of less than 5 ppm during the smaller Antarctic isotope maxima events as observed during the last glacial period (Ahn and Brook, 2014; Bereiter et al., 2012).To investigate how atmospheric CO2 is related with climate change on millennial time scales during MIS 6, we reconstructed 150 samples of atmospheric CO2 data from the EPICA Dome C (EDC) ice core during the MIS 6 period (189.4-135.4 kyr BP). One minor and five major variabilities of atmospheric CO2 during the early MIS 6 period (189-160 kyr BP) were found. These variabilities are highly matched with Antarctic temperature. During the short stadials in the North Atlantic, atmospheric CO2 variations are negligible and decoupled with temperature variations in Dome C. During this period, the strength of upwelling in the southern ocean might not be sufficient to impact on atmospheric CO2. In addition, 2 modes of CO2 variations are present in the MIS 6 period. Carbon dioxide maxima (CDM) 6 lags abrupt warming in the Northern Hemisphere by only 100±360 yrs, while the lags for CDM 3 and 4 are much longer, 1,100±280 yrs on average. Theses 2 modes of CO2 variations might be related with a mode change of AMOC from the earliest MIS 6 to MIS 6.5. These two phenomena also are observed during the last glacial period. However, the limited available proxy data permit only an exploratory discussion of the mechanisms responsible for CO2 variability during MIS6. Because the boundary conditions of the last glacial period cannot be applied to MIS 6, additional proxy data and multiple modelling studies conducted during MIS 6 period are needed.