Our main objective in this paper is to propose an approach to allocate variable input costs among crops produced by farmers based on panel data including information on input expenditure aggregated at the farm level. Our proposed approach simultaneously allows to control for unobserved farms and farmers heterogeneity, to address the potential dependence between variable input uses and acreage choice decisions, and to ensure consistent values of input use estimates. These are indeed major issues generally encountered in the estimation input allocation equations. This approach relies on a model of input allocation derived from accounting identities, in which unobserved input uses per crop are modeled as time-varying random parameters. Our model is estimated, using an extension of the Stochastic Approximation Expectation Maximization algorithm, on a sample of French farms’ accounting data. Our estimation results show good performance of our approach compared to standard regression approaches generally used by agricultural economists.