During recent years, new on-line methods in mass spectrometry have been developed for measuring Δ 18O in organic material. They allow a much higher sample throughput than off-line methods with the result that sample preparation becomes the time-limiting factor. Therefore we tested whether analysis of tree ring samples can be made on whole wood instead of pure cellulose, which until now was commonly used in almost all 18O tree ring studies. Measurements with an on-line method based on pyrolysis in an elemental analyzer show that the tree ring Δ 18O time series of wood and cellulose from an oak of the Swiss Mittelland are similar (r2=0.65). However, there are significant differences, and some climatic information may be lost if bulk wood is analyzed instead of cellulose. This can partly be balanced by an increased sample throughput, resulting in the averaging of more data. Further, we improved an off-line method (pyrolysis in a nickel tube followed by catalytic CO to CO2 conversion on nickel powder) by adding a CO2 trap to enhance the CO to CO2 conversion. The best reproducibility associated with this method is better than 0.1%. We also found a strong memory effect linked with this method, causing a dampening of the signal of 30--40%. Therefore published climatic interpretation of Δ 18O data measured using similar methods may require revision. ¿ 1999 American Geophysical Union