There is an increasing need to understand the pre-Quaternary warm climates, how climate–vegetation interactions functioned in the past, and how we can use this information to understand the present. Here we report vegetation modelling results for the Late Miocene (11–7 Ma) to study the mechanisms of vegetation dynamics and the role of different forcing factors that influence the spatial patterns of vegetation coverage. One of the key uncertainties is the atmospheric concentration of CO2 during past climates. Estimates for the last 20 million years range from 280 to 500 ppm. We simulated Late Miocene vegetation using two plausible CO2 concentrations, 280 ppm CO2 and 450 ppm CO2, with a dynamic global vegetation model (LPJ-GUESS) driven by climate input from a coupled AOGCM (Atmosphere-Ocean General Circulation Model). The simulated vegetation was compared to existing plant fossil data for the whole Northern Hemisphere. For the comparison we developed a novel approach that uses information of the relative dominance of different plant functional types (PFTs) in the palaeobotanical data to provide a quantitative estimate of the agreement between the simulated and reconstructed vegetation. Based on this quantitative assessment we find that pre-industrial CO2 levels are largely consistent with the presence of seasonal temperate forests in Europe (suggested by fossil data) and open vegetation in North America (suggested by multiple lines of evidence). This suggests that during the Late Miocene the CO2 levels have been relatively low, or that other factors that are not included in the models maintained the seasonal temperate forests and open vegetation.