Ground-based observations of land–atmosphere fluxes are necessary to progressively improve global climate models. Observed data can be used for model evaluation and to develop or tune process models. In arctic permafrost regions, climate–carbon feedbacks are amplified. Therefore, increased efforts to better represent these regions in global climate models have been made in recent years. We present a multi-annual time series of land–atmosphere carbon dioxide fluxes measured in situ with the eddy covariance technique in the Siberian Arctic (72∘22′ N, 126∘30′ E). The site is part of the international network of eddy covariance flux observation stations (FLUXNET; site ID: Ru-Sam). The data set includes consistently processed fluxes based on concentration measurements of closed-path and open-path gas analyzers. With parallel records from both sensor types, we were able to apply a site-specific correction to open-path fluxes. This correction is necessary due to a deterioration of data, caused by heat generated by the electronics of open-path gas analyzers. Parameterizing this correction for subperiods of distinct sensor setups yielded good agreement between open- and closed-path fluxes. We compiled a long-term (2002 to 2017) carbon dioxide flux time series that we additionally gap-filled with a standardized approach. The data set was uploaded to the Pangaea database and can be accessed through https://doi.org/10.1594/PANGAEA.892751.