The Earth’s mantle holds more carbon than its oceans, atmosphere and continents combined, yet the distribution of carbon within the mantle remains uncertain. Global variations in the carbon content of the depleted mantle have been inferred from carbon-trace element systematics of ultra-depleted mid-ocean ridge glasses and melt inclusions, though the origin of mantle carbon variability remains uncertain. A variety of observations suggest that the mantle plumes underlying ocean islands have high bulk carbon contents, but the characterisation of individual mantle plume components has not been possible.We supplement the existing melt inclusion data from Iceland with four new datasets, significantly en- hancing the spatial and geochemical coverage. Within the combined dataset there is significant variation in melt inclusion CO2/Ba ratio, which is tightly correlated with trace element enrichment. The trends in CO2/Ba–Ba space displayed by our new data coincide with the same trends in data compiled from global ocean islands and mid-ocean ridges, forming a global array. The overall structure of the global CO2/Ba array is not a property of the source, instead controlled by CO2 degassing before melt inclusion entrapment, and CO2 loss by decrepitation following inclusion entrapment.The position of melt inclusion datasets within the global array allows us to filter out inclusions likely tohave undergone decrepitation. Providing earlier CO2 degassing and magma mixing have not destroyed themantle signal, the remaining inclusions may preserve mantle CO2/Ba values. Using the observed covariationof the filtered CO2/Ba ratio with radiogenic isotope tracers of mantle source, we find CO2 concentrationsin the Icelandic depleted mantle of 102+53 -27 ppmw, typical of mantle underlying the mid-ocean ridge system,and high CO concentrations of 2.2+3.8 -1.5 wt% associated with the high 3He/4He component of the Iceland plume. However, source and process are difficult to deconvolve for the eruptions most sensitive to enriched, or recycled, mantle components; the presently available data is consistent with recycled material being either enriched or depleted in carbon.Extending our approach globally, we find no systematic variation in CO2/Ba or CO2/Nb with enrichment (in the absence of a primordial mantle contribution). The lack of global co-variation suggests mechanisms recycling CO2 from the surface into the convecting mantle are decoupled from those recycling most trace elements. However, the variation of CO2/Nb with Ba/Nb suggests Ba may behave similarly to CO2 during recycling.