The potential of field hyperspectral remote sensing data for non-destructive assessment of hay meadow biomass and vascular plant diversity has been investigated. Spectrometric and agronomic data were acquired at peak biomass over 34 sites distributed at diverse elevation and slopes over an area of 220 km2 in the Central Alps (Valtellina, Northern Italy). Different modelling approaches were tested to evaluate the predictive performance of spectral measurements: (i) the use of two band ratios of reflectance as input in ordinary least square regression models and (ii) the use of all reflectance bands as input in multivariate partial least square regression models. Each model was subjected to leave-one-out cross-validation and evaluated using the cross-validated coefficient of determination and the root mean square error. Fresh biomass and fuel moisture content were predicted with an average error of <20%, while Shannon Diversity Index and plant species richness were predicted with an average error of <15%, with no relevant differences between the two modelling approaches. Best models for plant diversity indicator prediction were based on chlorophyll/nitrogen sensible bands in the blue and red spectral regions. This observation together with the apparent negative correlation between hay meadow plant diversity and canopy chlorophyll/nitrogen content in the study area suggests a potential connection between reflectance and plant diversity indicators based on meadow biochemical properties.