Heterotrophic respiration (Rh) is, at a global scale, one of the largest CO2 fluxes between the Earth's surface and atmosphere and may increase in the future. The previous generation of Earth system models (ESMs) was able to reproduce global fluxes relatively well, but at that, time no gridded products were available to perform an in-depth evaluation. The capacity of the new generation of ESMs used within the Coupled Model Intercomparison Project Phase 6 (CMIP6) to reproduce this flux has not been evaluated, meaning that the realism of resulting CO2 flux estimates is unclear. In this study, we combine recently released observational data on Rh and ESM simulations to evaluate the ability of 13 ESMs from CMIP6 to reproduce Rh. Only 4 of the 13 tested ESMs were able to reproduce the total Rh flux, but spatial analysis underlined important bias compensation for most of the ESMs, which generally showed an overestimation in tropical regions and an underestimation in arid regions. To identify the main drivers of the bias, we performed an analysis of the residuals and found that mean annual precipitation was the most important driver explaining the difference between ESM simulations and observation-derived products of Rh, with a higher bias between ESM simulations and Rh products where precipitation was high. Based on our results, next-generation ESMs should focus on improving the response of Rh to soil moisture.