Abstract. Critical data selection is essential for determining representative baseline levels of atmospheric trace gas measurements even at remote measuring sites. Different data selection techniques have been used around the world which could potentially lead to bias when comparing data from different stations. This paper presents a novel statistical data selection method based on CO2 diurnal pattern occurring typically at high elevated mountain stations. Its capability and applicability was studied for atmospheric measuring records of CO2 from 2010 to 2016 at six Global Atmosphere Watch (GAW) stations in Europe, namely Zugspitze-Schneefernerhaus (Germany), Sonnblick (Austria), Jungfraujoch (Switzerland), Izaña (Spain), Schauinsland (Germany) and Hohenpeissenberg (Germany). Three other frequently applied statistical data selection methods were implemented for comparison. Among all selection routines, the new method named Adaptive Baseline Finder (ABF) resulted in lower selection percentages with lower maxima during winter and higher minima during summer in the selected data. To investigate long-term trend and seasonality, seasonal decomposition technique STL was applied. Compared with the unselected data, mean annual growth rates of all selected data sets were not significantly different except for Schauinsland. However, clear differences were found in the annual amplitudes as well as for the seasonal time structure. Based on correlation analysis, results by ABF selection showed a better representation of the lower free tropospheric conditions.