Semi-natural grassland biotopes represent a specific feature of the Poloniny National Park (East Carpathian Mts, Slovakia) with the great biodiversity values. In this article we use habitat models to predict their distribution in the study area. Based on 285 phytosociological records, we identified 5 main grassland types: mountain meadows, mesophilous meadows, wet meadows, dry meadows and fens. We used logistic regression and classification and regression trees (CART) for creating habitat models for each respective grassland type. Basic environmental variables derived from digital elevation model (elevation, slope, aspect, topographical wetness index, potential direct solar irradiation), landscape structure (landscape indexes: number of different classes, diversity, fragmentation), soil map (fertility index, soil type) and geomorphology were analysed as predictors. In general, the highest predictive power was shown by elevation, slope, topographic wetness index and geomorphologic form. The main output of habitat modelling was the map of potential grassland types of the whole study area. Final map of predicted distribution of grasslands emerged after the intersection of grassland landcover class with the map of potential grassland types. The accuracy was assessed comparing the modelled distribution of grassland types to the validation subset (approx. 30% of original dataset). Generally, logistic-based model exhibited better accuracy. Its results have shown that almost 81% of all semi-natural grasslands in the study area represent mesophilous meadows followed by wet meadows (13%), fens (2.5%), dry meadows (2%) and mountain meadows (1.5%).