Over the past 10 years, there has been a growing debate about the relative importance of late-Quaternary variations in climate and CO₂ as drivers of changes in vegetation structure. In this study, we compare new datasets of pollen-reconstructed leaf area index values (LAIs) for northern and eastern North America with BIOME4 simulations and regional vegetation histories in order to identify sources of data-model disagreements and assess the relative influence Of CO₂ and climate on LAI. BIOME4, an equilibrium biogeochemistry-biogeography vegetation model, was forced by climate simulations from the Hadley Centre Unified GCM to simulate changes in LAI at 1000-year intervals for the past 21,000 years in six regions of North America. BIOME4 simulations included sensitivity experiments to investigate the effects Of CO₂ and climate on LAI. Pollen-reconstructed LAIs were based on the modern analog technique and a cross-referencing of surface pollen samples against recent LAI observations from the MODIS sensor. Pollen-reconstructed LAIs were biased towards too high values due to lower-than-present atmospheric CO₂; sensitivity experiments with BIOME4 suggested that this bias ranged from 13±7% during the Lateglacial period (21-12 ka) to 7±4% during the Holocene (11-0 ka). Simulated and reconstructed LAI trends agreed well for western Canada, southeastern Canada/northern US, and eastern North America, but significant discrepancies were observed in the Midwestern LIS, Alaska, and southeastern US. Causes of data-model discrepancies varied by region and were due to (1) inaccurate simulation of the prairie-forest ecotone in the Midwestern US due to too-wet climate simulations; (2) a combination of too-high simulated LAIs for Lateglacial tundra and too-low reconstructed LAIs from Alaskan pollen samples that lacked modern analogs; and (3) too-high reconstructed LAIs for scrublands in the southeastern US. Sensitivity tests indicated that late-Quaternary changes in LAI were driven primarily by climate, with CO₂ as a secondary influence. Sensitivity simulations in which climate was allowed to vary while CO₂ was constant agreed better with reconstructed LAIs and regional vegetation histories than did the simulations in which CO₂ varied and climate was constant. Our work emphasizes the need for models to incorporate the effects of both CO₂ and climate on late-Quaternary vegetation dynamics and structure. We show that LAI is a viable paleo-vegetation proxy that can be used as a benchmark for rigorous, quantitative data-model comparisons focusing primarily on vegetation history, particularly in regions with high densities of fossil pollen data.