The risk associated to the production of biomass sorghum (Sorghum bicolor (L.) Moench) to feed a power plant in the Po Valley (Italy) was studied with a modelling approach. Available biomass was modelled by CropSyst, coupled to a "sorghum haying model", using three sorghum genotypes, of contrasting earliness (early, medium-late and late), on a mosaic of virtual farms created in the target cropping area. The energy performance, from cradle to farm gate, of the biomass production system was performed calculating Energy Return on Investment (EROI), Net Energy Gain (NEG) and Energy Use Efficiency (EUE). The highest baled biomass (14.0 Mg DM ha(-1)y(-1)) was obtained with the early maturity type that had less haymaking failures (6.9%), followed by the late and medium-late genotypes. As a consequence, the early maturity type had the highest probability of exceeding the biomass needs of the power plant on a cropping area of 4222 ha. The early genotype also had the highest EROI (14.8) and NEG (205.6 GJ ha(-1)y(-1)) and the lowest EUE (1.06 GJ Mg-1 DM y(-1)). To achieve a 0.5 probability to exceed the target biomass production, the area to be cultivated should be 4558, 5160 and 4962 ha for the early, medium-late and late genotypes, respectively. (C) 2016 Elsevier Ltd. All rights reserved.