The carbon isotopic composition of pedogenic carbonate (delta 13Cc) is a widely used proxy for paleoatmospheric CO2 levels. However, the estimated CO2 is often highly uncertain due to the lack of constraints on other variables that influence this system. To improve paleosol-based CO2 reconstructions, here we propose the Joint Proxy Inversion (JPI) of a paleosol-based proxy system model, which integrates various proxies (e.g., pedogenic carbonate delta 13C and delta 18O) that share sensitivities to common environmental parameters (e.g., CO2, precipitation) via a forward proxy model in a Bayesian hierarchical framework. JPI generates self-consistent posterior distributions of environmental parameters conditioned on multi-proxy data sets while accounting for proxy and model uncertainties. The JPI was applied to data from Quaternary loess deposits from the Chinese Loess Plateau to test its performance. Sensitivity tests and posterior CO2 estimates suggest insensitivity of proxy values to CO2 when the analysis integrates delta 13C, delta 18Oc, and clumped isotope values of pedogenic carbonates only. Improved results were achieved after incorporating a precipitation-sensitive proxy, a model component representing environmental variation as a function of time, and a parameterized relationship between global temperature and CO2. The improved model performance is indicated by agreement between posteriors of environmental parameters (i.e., CO2, temperature, precipitation) and other independent records. Compared to the traditional proxy interpretation approach, the paleosol-JPI offers a more holistic and rigorous way to interpret multi-proxy records with shared environmental sensitivities, provides additional information on climate and environmental systems, and is capable of reconstructing paleo-CO2 levels with higher precision and fidelity.