Historical reconstructions of plant community distributions are useful for biogeographic studies and restoration planning, but the quality of insights gained depends on the depth and reliability of historical information available. For the Central Valley of California, one of the most altered terrestrial ecosystems on the planet, this task is particularly difficult given poor historical documentation and sparse relict assemblages of pre-invasion plant species. Coastal and interior prairies were long assumed to have been dominated by perennial bunchgrasses, but this hypothesis has recently been challenged. We evaluated this hypothesis by creating species distribution models (SDMs) using a novel approach based on the abundance of soil phytoliths (microscopic particles of biogenic silica used as a proxy for long-term grass presence) extracted from soil samples at locations statewide. Modeled historical grass abundance was consistently high along the coast and to a lesser extent in higher elevation foothills surrounding the Central Valley. SDMs found strong associations with mean temperature, temperature variability, and precipitation variability, with higher predicted abundance in regions with cooler, equable temperatures and moderated rainfall, mirroring the pattern for modern perennial grass distribution across the state. The results of this study strongly suggest that the pre-Columbian Central Valley of California was not dominated by grasses. Using soil phytolith data as input for SDMs is a promising new method for predicting the extent of prehistoric grass distributions where alternative historical datasets are lacking.