We present a dual cost function estimation for total farm cost in a programming model setup, with individual crop shares and expected yields as arguments, estimated simultaneously with risk behaviour. Using large unbalanced samples of specialized arable farms from Northern Italy, the French Grandes Culture Region and Cologne-Aachen in Germany that are observed for at least three consecutive years over the time period 1995-2008, we find a quite satisfactory fit for crop shares and total costs. We implement two model variants where zero crop observations are considered only in the second variant. Our results indicate that the specialized arable crop farmers in the samples use crop shares only to a limited degree as an instrument of risk management. We find moderate technical progress and large efficiency differences between farms.