The identification of regional representative carbon monoxide (CO) measurements that are minimally influenced by local sources/sinks is essential to understand the characteristics of atmospheric CO over a certain region. In this study, three commonly used data filtering approaches were applied to atmospheric CO data obtained from 2010/2011 to 2017 at two World Meteorological Administration/Global Atmospheric Programme (WMO/GAW) regional stations (Lin'an, LAN and Shangdianzi, SDZ) in China, to study their applicability for individual stations. The three methods used were the meteorological conditions (MET), statistical approaches (robust extraction of baseline signal, REBS), and the time scale of the CO variations (standard deviations of the running mean, SDM). The results from the three methods displayed almost the same seasonal cycles at LAN but different variations at SDZ. They each extracted similar yearly CO growth rates at LAN, but there was a large difference at SDZ, with values of -10.6 ± 0.5, -2.2 ± 0.1, and - 23.5 ± 0.3 ppb yr-1 for MET, REBS, and SDM, respectively. The slight decrease observed using REBS at SDZ was mainly due to the biased distribution of CO records, which was a purely statistical method that did not consider topography or meteorological conditions. Thus, the REBS method should be applied cautiously to CO observations at stations like SDZ. The SDM method may overestimate multi-year trends. Among the three approaches, MET may be the most suitable for filtering CO observation records, especially at stations like SDZ with special geographical and meteorological conditions in economically-developed regions.