Mineral dust is an important component of the Earth system due to its role in oceanic nutrient supply, cloud formation and its radiative properties. Changes in transport pathways and fluxes of mineral dust have attracted increased attention using radiogenic isotope analysis for detailed investigation of changing dust sources through time. However, multi-isotope studies provide complex datasets of dust provenance, often without exact quantification of source contributions. Here we use Bayesian mixing models and existing radiogenic isotope data to quantify changes in South Pacific dust provenance for the Holocene and the Last Glacial Maximum (LGM; ∼18–24 ka BP). Testing different model configurations showed grouping small source regions to single continental scale end members prior to modeling can lead to biased results, and so we group model outputs post-modeling. During the LGM, a higher proportion (mean 53%) of dust entering the South Pacific was South American in origin, compared to a Holocene mean of 31%. In contrast, Australian dust contributions were lower during the LGM (mean 38%) than Holocene (mean 55%), with significant spatial gradients for both time slices. In the subpolar South Pacific, the high representation of South American dust during the LGM (up to ∼75%) coincides with larger dust particles; together indicating that far-traveled dust transport was facilitated by long atmospheric residence times and an accelerated westerly wind circulation during this time. Our study shows how Bayesian mixing models provide valuable constraints for dust source contributions, an approach which may help in the calibration of atmospheric models, using complex isotopic datasets.