We describe and implement a data selection algorithm for identifying background observations of atmospheric CO2 from continuous measurements in situ. Several data selection criteria have been developed and are currently used: the main objective of this work is to define a common methodology for extracting the atmospheric background signal while minimizing heterogeneities caused by the use of different selection algorithms. The algorithm used in this study (BaDS, Background Data Selection) was tested and optimized using data (from 2014 to 2018) from four Italian stations featuring markedly different environmental conditions (e.g. mountainous, coastal and marine): Plateau Rosa (PRS), Mt. Cimone (CMN), Capo Granitola (CGR) and Lampedusa (LMP). Their locations extend from the Alps to the central Mediterranean. The adopted algorithm proved to be effective in separating the local/regional signal from the background signal in the CO2 time series. Approximately 6% of the LMP data, 11% of the PRS, 20-38% of the CMN, and 65% of the CGR were identified as data not attributable to the atmospheric background. LMP and PRS can be used as reference sites for the central Mediterranean area, while CMN and CGR were more influenced by regional sources and sinks. Finally, we discuss a possible application of the methodology at other measurement stations.