Radiocarbon (C-14) analysis on CO2 can provide valuable information on the carbon cycle as different carbon pools differ in their C-14 signature. While fresh, biogenic carbon shows atmospheric C-14 concentrations, fossil carbon is C-14 free. As shown in previous studies, CO2 can be collected for C-14 analysis using molecular sieve cartridges (MSC). These devices have previously been made of plastic and glass, which can easily be damaged during transport. We thus constructed a robust MSC suitable for field application under tough conditions or in remote areas, which is entirely made of stainless steel. The new MSC should also be tight over several months to allow long sampling campaigns and transport times, which was proven by a one year storage test. The reliability of the (CO2)-C-14 results obtained with the MSC was evaluated by detailed tests of different procedures to clean the molecular sieve (zeolite type 13X) and for the adsorption and desorption of CO2 from the zeolite using a vacuum rig. We show that the (CO2)-C-14 results are not affected by any contamination of modern or fossil origin, cross contamination from previous samples, and by carbon isotopic fractionation. In addition, we evaluated the direct CO2 transfer from the MSC into the automatic graphitization equipment AGE with the subsequent C-14 AMS analysis as graphite. This semi-automatic approach can be fully automated in the future, which would allow a high sample throughput. We obtained very promising, low blank values between 0.0018 and 0.0028 (FC)-C-14 (equivalent to 50,800 and 47,200 yrs BP), which are within the analytical background and lower than results obtained in previous studies. (C) 2017 Elsevier B.V. All rights reserved.