Estimating greenhouse gas (GHG) balances of biofuels is subject to significant biases stemming from modelling choices about system definition and boundaries, functional unit, reference systems and allocation methods. This paper investigates extent to which these choices influence the results. After performing a comparison and constructive criticism of various modelling choices, life cycle assessment (LCA) of wheat-to-bioethanol was used as an illustrative case where bioethanol was blended with gasoline at various percentages (E5, E10 and E85). A large difference was observed in reduction of GHG emissions with a high sensitivity to method used to allocate impacts between co-products, type of reference systems, choice of functional unit and type of blend. The study recommends for basing estimation of energy and GHG balances of biofuels on principles of transparency, consistency and accuracy.