Methods are described for measuring changes in atmospheric O2 concentration with emphasis on gas handling procedures. Cryogenically dried air samples are collected in 5 L glass flasks at ambient pressure and analyzed against reference gases derived from high-pressure aluminum tanks. Fractionation effects are minimized by avoiding pressure and flow variations throughout the gas-handling system. The overall external reproducibility is approximately ¿3.3 per meg, with systematic errors associated with collecting samples and with storing them for 1 year reduced to the level of 3 per meg or smaller. The demonstrated short-term reproducibly of air delivered from high-pressure tanks is ¿1.5 per meg, with the composition changing by at most 5 per meg by surface desorption reactions as the tank is depleted to below 3500 kPa. A 9-year survey of a suite of six reference gases showed no systematic long-term trends in relative O2 concentrations to the level of 5 per meg. Results are presented from samples collected at Cape Grim (41 ¿S), Macquarie Island (54 ¿S) and the South Pole Station (90 ¿S). From measurements spanning 1991--1995 it is found that the O2 concentrations at the South Pole are on average 3.6¿1.2 per meg higher than at Cape Grim. This result runs contrary to the expectation that the air at high southern latitudes should be depleted in O2 as a result of O2 uptake from the Southern Ocean and may require the existence of unknown O2 sources near Antarctica or unexpected atmospheric transport patterns. ¿ 1998 American Geophysical Union