The aim of this thesis was to provide insight into past human activities in Amazonia, and the long-term forest recovery that followed, using phytolith analysis. The first part of this thesis focused on improving phytolith analysis as a proxy for vegetation changes in Amazonia. The morphological variation of 24 Amazonian palm phytoliths were assessed (Chapter 2) to improve the taxonomic resolution of palm phytolith identification, and results indicate the potential to differentiate Euterpe, Bactris, Oenocarpus, Attalea, Iriartea deltoidea, and Socratea exorrhiza. Chapter 3 demonstrated that phytolith assemblages (from terrestrial soil cores) varied across a gradient of (modern) human disturbance in Surinamese rainforests. In Chapters 4-6, we developed beta regression and GLM models to predict forest cover and biomass changes within 200 m and 1 km of Amazonian lakes, respectively, using grass phytoliths. Applying these innovations in Chapters 5-6 demonstrated that past human activities were on localized scales in Suriname and temporally heterogenous in Ecuador. Palm enrichment of Attalea, Oenocarpus and Astrocaryum occurred within 0 km, 1 km, and 8 km of an archaeological site in Suriname (Chapter 5). In Ecuador, forest cover and biomass ranged between 48-84% and 77-247 Mg/ha, respectively, and the largest decreases (between 1000-1255 CE) were paired with fires (Chapter 6). Overall, the type, intensity, timing, and frequency of disturbances are important factors influencing long-term forest recovery and ecological legacies in Amazonia.