In order to enhance the effectiveness of comparisons between modelled and empirical data for present and past vegetation, it is important to improve the characterization of tropical grass-dominated biomes reconstructed from fossil tracers. This study presents a method for assigning phytolith assemblages to tropical grass-dominated biomes, with the objective of offering a new tool for combining pollen and phytolith data in the reconstruction of tropical biomes. The West African latitudinal transect studied here extends from 12° N (southern Senegal) to 23° N (southern Mauritania), passing through the Guinean, Sudanian, Sahelian and Saharan bioclimatic zones. Modern phytolith assemblages were extracted from 59 soil surface samples taken throughout the study area and allocated, a priori, to three current biomes: (1) desert C₄ grassland, (2) short grass savanna, and (3) tall grass savanna. Five out of nine phytolith types identified were used as predictors in a discriminant analysis (with calibration and validation steps) for assigning phytolith assemblages to biomes. In addition, 74 modern pollen spectra from the West African transect, acquired from the African Pollen Database ( http://medias.obs-mip.fr/apd), were processed by the biomization method. This mathematical procedure involves assigning palynological taxa to one or more plant functional types, which represent broad classes of plants. The plant functional types, in turn, are combined to define biomes following a specific set of algorithms and rules. The resulting maps of the phytolith biomes thus derived were compared with maps of pollen biomes and of contemporary ecosystem classes. In the calibration and validation steps, 91.5% and up to 83%, respectively, of the phytolith samples were assigned to the correct biome. The short grass savanna and tall grass savanna biomes were assigned with similar accuracy by both the phytolith and pollen biomization methods, but the phytolith method gave substantially superior results for the desert C₄ grassland biome, providing seven out of seven correct assignments, compared with just one out of four by pollen biomization. Comparisons between an existing ecosystem map and the maps created from phytolith estimation showed close correspondence for desert C₄ grassland, short grass savanna and tall grass savanna, the latter providing correct assignments in 88, 62 and 91% of cases, respectively. The phytolith discriminant analysis method presented here accurately estimates three C₄ grass-dominated biomes that are widespread in West Africa. Complementarities between the phytolith method and pollen biomization are highlighted. Combining complementary phytolith and pollen data would provide more accurate assignments of C₄ grass-dominated biomes than pollen biomization alone.