This paper contributes to the debate on the nutrition-related outcomes of cash crop adoption by using a model with essential heterogeneity and a semi-parametric estimation technique. The model explicitly frames non-separability between production and consumption decisions of farming households providing an original test of separability. The empirical application is run using Malawian data. The results imply rational anticipations and decision process of agrarian households relative to the crop portfolio choice, disparate strength of market barriers faced by the farmers, non-separability between production and consumption decisions and a weak transmission from agricultural incomes to higher food expenditures and better diet.