Unpredicted, rapid plume elongation has been observed at subsurface CO 2 storage projects worldwide, exemplified by the Sleipner project. We show that conventionally ignored centimeter-meter scale heterogeneity in capillary pressure characteristics can manifest as rapid field-scale, decameter-kilometer, plume migration. We analyze the effect in the Goldeneye field, UK, a proposed storage site with a unique combination of sample/data accessibility and generality as an archetype sandstone reservoir. We overcome previous barriers by characterizing in greater detail over larger scales-the 65 m reservoir height at cm-m resolution-and through use of an upscaling scheme which resolves small-scale heterogeneity impacts in field-scale simulations. These models reveal that significant early time retardation of buoyantly rising CO 2 plumes is followed by rapid migration under the caprock in the presence of anisotropic, layered heterogeneities. Lateral migration speeds can be enhanced by 200%, placing first-order controls on fluid flow and providing a mechanistic explanation for field observations. Plain Language Summary Geological carbon storage is a promising technique to reduce greenhouse gas emissions. Captured carbon dioxide is generally injected into a subsurface reservoir over 1,000 m underground, displacing resident brine and eventually becoming trapped underneath a low-permeability caprock seal. However, at several industrial-scale storage sites around the world, the carbon dioxide has migrated laterally away from the injection well much quicker than anticipated and followed pathways that are not predicted by models. It is crucial that these models can predict the migration and demonstrate safe storage to owners and policy makers. In this work, we show that one source of the discrepancy is the omission of the impacts of small-scale rock heterogeneities in these models. We experimentally characterize rock cores from a North Sea reservoir at high resolution, and through rigorous multiscale modeling show that centimeter-meter-scale heterogeneities in the rock structure, for example, small mudstone layers in sandstone, can cause rapid migration at larger, meter-kilometer scales. Carbon dioxide can migrate up to 200% faster in the presence of layered heterogeneities. These heterogeneities are ubiquitous in nature and provide an explanation for the behavior seen at storage sites worldwide. Our modeling approach incorporates this behavior, improving the predictability and control of storage operations.