This paper proposes the use of the Generalized Maximum Entropy (GME) method to estimate input allocation in multi-crop systems using heterogeneous data sources (farm accountancy data and cropping practices survey data). The aim is to explore the role of well-defined a priori information in improving the accuracy of GME estimation. The performance of the GME method is compared afterward to a Bayesian approach— Highest Posterior Density (HPD)—to assess their accuracy when reliable non-sample (prior) information is used and investigate their usefulness for reconciling heterogeneous data sources. Both approaches are applied to a given set of farm accounting data which reports information on input allocation between alternative input uses. The estimation results show that the use of well-defined prior information from external data source improves GME estimates even though this performance is not always significant. It also appears that the Bayesian (HPD) approach could be a good alternative to the GME estimator. HPD provides results that are close to the GME method with the advantage of a straightforward and transparent implementation of the a priori information.