Identification of ecosystem turnover in the palaeo-vegetation record is important for understanding the resilience of ecosystems to past environmental change. There is uncertainty over the ability of different types of palaeo-vegetation proxies to detect ecosystem turnover. The aim of this paper is to compare the sensitivity of two palaeo-vegetation proxies – pollen and phytoliths – to changes within and between three key tropical South American ecosystems: evergreen forest, dry forest and savannah. A quantitative approach is used to assess the sensitivity of these two proxies to vegetation changes, based on the variability of proxy assemblages from 1-ha ecological plots in ecotonal south-west Amazonia. This modern dataset of proxy variability within evergreen forest, dry forest and savannah plots is then used to define thresholds for proxy variability which differentiate floristic changes within an ecosystem from ecosystem turnover. These thresholds are applied to two palaeo-vegetation records from NE Bolivia. Our results show that pollen is more sensitive than phytoliths to changes within evergreen forest, but phytoliths are more sensitive than pollen to changes within dry forest. Both proxies were equally sensitive to changes within the savannahs. These are important considerations for palaeoecologists selecting proxies for the study of ecosystem turnover in the palaeo-record. Application of the thresholds to the palaeo-record demonstrated the utility of this quantitative approach for assessing the magnitude of vegetation change in the palaeo-record. This quantitative approach is therefore a useful tool to improve the identification of ecosystem turnover in the palaeo-record.