Atmospheric transport models are used to constrain sources and sinks of carbon dioxide requiring that the modeled spatial and temporal concentration patterns are consistent with the observations. Serious obstacles to this approach are the sparsity of sampling sites and the lack of temporal continuity among observations at different locations. A procedure is presented that attempts to extend the knowledge gained during a limited period of measurements beyond the period itself resulting in records containing measurement data and extrapolated and interpolated values. From limited measurements we can define trace gas climatologies that describe average seasonal cycles, trends, and changes in trends at individual sampling sites. A comparison of the site climatologies with a reference defined over a much longer period of time constitutes the framework used in the development of the data extension procedure. Two extension methods are described. The benchmark trend method uses a deseasonalized long-term trend from a single site as a reference to individual site climatologies. The latitude reference method utilizes measurements from many sites in constructing a reference to the climatologies. Both methods are evaluated and the advantages and limitations of each are discussed. Data extension is not based on any atmospheric models but entirely on the data themselves. The methods described here are relatively straightforward and reproducible and result in extended records that are model independent. The cooperative air sampling network maintained by the National Oceanic and Atmospheric Administration Climate Monitoring and Diagnostics Laboratory in Boulder, Colorado, provides a test bed for the development of the data extension method; we intend to integrate and extend CO2 measurement records from other laboratories providing a globally consistent atmospheric CO2 database to the modeling community. ¿ American Geophysical Union 1995