The seasonality in cave CO₂ levels was studied based on (1) a new data set from the dynamically ventilated Comblain-au-Pont Cave (Dinant Karst Basin, Belgium), (2) archive data from Moravian Karst caves, and (3) published data from caves worldwide. A simplified dynamic model was proposed for testing the effect of all conceivable CO₂ fluxes on cave CO₂ levels. Considering generally accepted fluxes, i.e., the direct diffusive flux from soils/epikarst, the indirect flux derived from dripwater degassing, and the input/output fluxes linked to cave ventilation, gives the cave CO₂ level maxima of 1.9?×?10^?2 mol m^?3 (i.e., ～ 440 ppmv), which only slightly exceed external values. This indicates that an additional input CO₂ flux is necessary for reaching usual cave CO₂ level maxima. The modeling indicates that the additional flux could be a convective advective CO₂ flux from soil/epikarst driven by airflow (cave ventilation) and enhanced soil/epikarstic CO₂ concentrations. Such flux reaching up to 170 mol s^?1 is capable of providing the cave CO₂ level maxima up to 3?×?10^?2 mol m^?3 (70,000 ppmv). This value corresponds to the maxima known from caves worldwide. Based on cave geometry, three types of dynamic caves were distinguished: (1) the caves with the advective CO₂ flux from soil/epikarst at downward airflow ventilation mode, (2) the caves with the advective soil/epikarstic flux at upward airflow ventilation mode, and (3) the caves without any soil/epikarstic advective flux. In addition to CO₂ seasonality, the model explains both the short-term and seasonal variations in δ¹³C in cave air CO₂.