We have developed in situ and flask sampling systems for airborne measurements of variations in the O2 /N2 ratio at the part per million level. We have deployed these instruments on a series of aircraft campaigns to measure the distribution of atmospheric O2 from 0–14 km and 87◦ N to 86◦ S throughout the seasonal cycle. The National Center for Atmospheric Research (NCAR) airborne oxygen instrument (AO2) uses a vacuum ultraviolet (VUV) absorption detector for O2 and also includes an infrared CO2 sensor. The VUV detector has a precision in 5 s of ±1.25 per meg (1σ ) δ(O2 /N2 ), but thermal fractionation and motion effects increase this to ±2.5–4.0 per meg when sampling ambient air in flight. The NCAR/Scripps airborne flask sampler (Medusa) collects 32 cryogenically dried air samples per flight under actively controlled flow and pressure conditions. For in situ or flask O2 measurements, fractionation and surface effects can be important at the required high levels of relative precision. We describe our sampling and measurement techniques and efforts to reduce potential biases. We also present a selection of observational results highlighting the individual and combined instrument performance. These include vertical profiles, O2 : CO2 correlations, and latitudinal cross sections reflecting the distinct influences of terrestrial photosynthesis, air–sea gas exchange, burning of various fuels, and stratospheric dynamics. When present, we have corrected the flask δ(O2 /N2 ) measurements for fractionation during sampling or analysis with the use of the concurrent δ(Ar/N2 ) measurements. We have also corrected the in situ δ(O2 /N2 ) measurements for inlet fractionation and humidity effects by comparison to the corrected flask values. A comparison of Ar/N2 -corrected Medusa flask δ(O2 /N2 ) measurements to regional Scripps O2 Program station observations shows no systematic biases over 10 recent campaigns (+0.2 ± 8.2 per meg, mean and standard deviation, n = 86). For AO2, after resolving sample drying and inlet fractionation biases previously on the order of 10–100 per meg, independent AO2 δ(O2 /N2 ) measurements over six more recent campaigns differ from coincident Medusa flask measurements by −0.3 ± 7.2 per meg (mean and standard deviation, n = 1361) with campaign-specific means ranging from −5 to +5 per meg.