<p>Gases preserved in ice cores provide a potential direct archive for atmospheric oxygen. Yet, oxygen-to-nitrogen ratios in ice cores (expressed as &#948;O<sub>2</sub>/N<sub>2</sub>) are modified by a number of processes related to gas trapping and gas losses in the ice. Such complications have long hindered the use of ice core &#948;O<sub>2</sub>/N<sub>2</sub> to derive true atmospheric oxygen concentrations. Recently, a persistent decline in &#948;O<sub>2</sub>/N<sub>2</sub>, observed in four different ice cores (GISP2, Vostok, Dome F, and EDC), is interpreted to reflect decreasing atmospheric O<sub>2</sub> concentrations over the late Pleistocene (Stolper et al., 2016). The rate of &#948;O<sub>2</sub>/N<sub>2</sub> change is -8.4&#177;0.2 &#8240;/Myr (1&#963;). Using new measurements made on EDC samples stored at -50 &#176;C and therefore free from gas loss, Extier et al (2018) confirms the decrease in &#948;O<sub>2</sub>/N<sub>2</sub> with a slope of -7.0&#177;0.6&#8240;/Myr (1&#963;).</p><p>Here, we present new &#948;O<sub>2</sub>/N<sub>2</sub> measurements made on 1.5-million-year-old blue ice cores from Allan Hills Blue Ice Areas, East Antarctica. We use argon-to-nitrogen ratios (&#948;Ar/N<sub>2</sub>) in the ice to correct for the fractionations during bubble close-off and gas losses. In those processes, &#948;Ar/N<sub>2</sub> is fractionated in a fashion similar to &#948;O<sub>2</sub>/N<sub>2</sub> (Huber et al., 2006; Severinghaus and Battle, 2006). Paired &#948;O<sub>2</sub>/N<sub>2</sub>-&#948;Ar/N<sub>2</sub> values measured from the same sample were classified into three different time slices: 1.5 Ma (million years old), 950 ka, and 490 ka. Between 950 ka and 490 ka, we observe a decline in &#948;O<sub>2</sub>/N<sub>2</sub> similar to that observed in the aforementioned deep ice cores. This observation gives us confidence in the validity of the Allan Hills blue ice &#948;O<sub>2</sub>/N<sub>2</sub> records. Between 1.5 Ma and 950 ka, however, there is no statistically significant trend in ice core &#948;O<sub>2</sub>/N<sub>2</sub>. Our results show a surprising lack of variability from 1.5 to 0.95 Ma; even during the past ~0.9 Ma, the rate of decline was very slow.</p>