A number of questions related to conservation biology and climate research require quantitative reconstructions of vegetation abundance or land-cover on a long-term historical perspective (e.g. Gaillard et al. 2010, Boyle et al., 2011, Kaplan et al., 2011). In particular, long-term changes in plant composition and abundance play a critical role in land coverclimate feedbacks (e.g. Wramneby et al. 2010) and, along with spatial distribution of species and habitats, they are probably one of the major factors affecting floristic and plant habitat diversity through time (e.g. Jeffers et al. 2011, Willis et al. 2010, Berglund et al., 2008, Willis and Birks, 2006). Jeffers et al. (2011) showed that abrupt changes in environmental conditions over time lead to similarly abrupt changes in tree-grass interaction outcomes. However, traditional percentage pollen records do not allow measuring the true extent of changes in e.g. the relationship between trees’ and herbs’ abundances/cover or other characteristics of plant spatial distributions through time (e.g. Gaillard et al., 2010). In the C3A project, questions related to the history of landscape fragmentation, and its consequences in terms of e.g. past changes in floristic diversity, require that the true abundance of the major plants composing the mountain forest and mixed forest-savanna landscapes of the past is quantified as precisely as possible before anything can be attempted in terms of reconstruction of past spatial distribution of plants or vegetation units, and changes in floristic diversity. In order to attempt a first quantification of Holocene changes in the abundance of major plant taxa of the mountain forest and savannas in Cameroon, we applied the Landscape Reconstruction Algorithm (LRA; Sugita 2007a and b) using five pollen records from contrasting vegetation regions today, i.e. one large (47 ha) and two small (8, and 25 ha) lakes in the savannas, and one large (198 ha) and one small (16.5 ha) bogs in the mountain forests. So far, the LRA was tested only in Europe using modern and historical pollen-vegetation datasets, and applied on fossil pollen records (e.g. Gaillard et al., 2010, Nielsen et al., 2011). The application of the LRA and its two models REVEALS (Regional Etimates of VEgetation Abundance from Large Sites) and LOVE (LOcal Vegetation Estimates) requires values of pollen productivity and fall speed of pollen (FSP) for the plant taxa to be reconstructed. Pollen productivity (PP) can be estimated using modern pollen data and related vegetation inventories and models of the pollen-vegetation relationship, i.e. the Extended R-Value model (ERV-model, e.g. Prentice and Parsons, 1983). Measurements of fall speed of pollen do exist for a number of temperate plant species, but none exist for tropical species. However, reasonable estimates can be obtained by the Stoke’s law on the basis of pollen grains’ mean size (e.g. Broström et al., 2008). FSP and PPs were calculated for only a few taxa of south African savannas (Duffin et al., 2008). We estimated PP for 16 major taxa on the basis of modern pollen data and related vegetation inventories at 21 sites in mountain forests and mixed forest-savannas. Preliminary FSPs for the 16 taxa were calculated on mean values of pollen sizes from the literature and few measures on slides from reference collections. First results indicate that, relatively to Gramineae (grasses), tree taxa such as Schefflera, Combretaceae and Syzygium have very high to high PPEs, while Asteraceae, Urticaceae, Podocarpus, Pavetta, and Bridelia have low to very low PPEs. Alchornea and Lophira have PPEs comparable to Gramineae. This implies that Schefflera will tend to be overrepresented while many other trees such as Podocarpus and Bridelia will often be underrepresented by pollen percentages. Poaceae will, as a consequence, be overrepresented compared to trees such as Podocarpus and Pavetta. We applied the LRA using the two large lakes to estimate past regional vegetation abundance (REVEALS model), and the three small sites to estimate the local vegetation abundance (LOVE model). Preliminary results show that the relevant source area of pollen (RSAP sensu Sugita 1994) of the three small sites varies with time between a 500 m -and 5000 m - radius area around the sites, which is the smallest area for which the LRA estimates of vegetation cover are reliable. The LRA reconstruction indicates that the sites located today in mixed forest-savanna were significantly more forested in the past than pollen percentages suggest, in particular before 3000 BP, but also 1500-2000 BP. Podocarpus and Alchornea in particular probably covered more land than pollen assemblages alone would suggest, while grasses covered much less land than its representation in pollen assemblages. Moreover, the change around 3000 BP is more spectacular when LRA estimates are considered. At the sites located in mountain forests today, the regional landscape was dominated by trees during most of the Holocene (REVEALS reconstruction at the large site), while tree pollen decreased in abundance around 3000 BP at the small site (LRA reconstruction). Podocarpus, Schefflera and Alchornea were the major tree pollen taxa. Before 3000 BP, Schefflera is dominant in the pollen assemblage, while Podocarpus is dominant in the LRA reconstruction. Alchornea is slightly better represented in the land cover than in the pollen assemblages. The LRA estimates indicate a later and more rapid increase of Gramineae from 3500 to 4000 BP than the pollen percentages (increase from 4000 BP). After 3000 BP, Schefflera is the dominant tree, although with less percentage cover than pollen percentages. The LRA reconstruction suggests that Podocarpus percentage cover dropped between 3000 and 2500 BP, but reached very low values later than indicated by the pollen percentages , i.e. ca. 2000 BP instead of 2500 BP. In conclusion, the major effect of the LRA correction is to 1) increase the tree/herb relationship before 3000 BP, 2) amplify the extent and rate of increase in Gramineae at the expense of trees around 3000 BP, and 3) inverse the Podocarpus/Schefflera relationship. New ERV-model runs to estimate PPs are in progress. They are based on a revised selection of 20 major taxa and on FSPs calculated from a large number of measurements on modern and fossil pollen grains. New LRA runs are planed with the objective to evaluate the first LRA reconstructions using a larger choice of FSPs and PPEs.