The H2O and CO2 contents of melt inclusions can potentially be used to infer pressures of crystallization and inclusion entrapment because the solubility of mixed H2O-CO2 vapor has been determined experimentally for a wide range of melt compositions. However, melt inclusions commonly develop a shrinkage bubble during post-entrapment cooling and crystallization because these processes cause a pressure drop in the inclusion. This pressure drop causes a vapor bubble to nucleate, leading to exsolution of low-solubility CO2 from the trapped melt. To investigate the loss of CO2 into such bubbles, we experimentally heated large, naturally glassy melt inclusions in olivine (Fo contents of 88.1 ± 0.2) from a Mauna Loa picrite to rehomogenize the inclusions. Rapid heating to 1420 °C using a high-temperature heating stage dissolved the shrinkage bubbles into the melt. CO2 contents measured by FTIR spectroscopy and recalculated for melt in equilibrium with the olivine host are 224–505 ppm (n = 11) for heated inclusions, much higher than the CO2 contents of naturally quenched inclusions from the same sample (38–158 ppm; n = 8). Pressures of inclusion entrapment calculated from the H2O and CO2 data for the heated inclusions range from 0.5 to 1.1 kbar, indicating that Mg-rich olivine crystallized at very shallow depths beneath the surface of Mauna Loa. Our results indicate that 40–90% (average 75%) of the original CO2 dissolved in the melt at the time of inclusion entrapment can be lost to the shrinkage bubble during post-entrapment cooling. We show that the computational method of Riker (2005), which predicts the pre-eruption shrinkage bubble size as a function of the difference between trapping temperature and pre-eruption temperature, successfully reproduces our experimental results. Our results demonstrate that the mass of CO2 contained in shrinkage bubbles must be considered to accurately infer original pressures of crystallization for melt inclusions. However, the effect is expected to be smaller for more H2O-rich melt inclusions than those studied here because the vapor bubble in such inclusions will have lower mole fractions of CO2 than the low-H2O inclusions in our study.