Scientific research generally follows two main methods: empirico-inductive (EI), gathering scattered, real-world qualitative/quantitative data to elaborate holistic theories, and the hypothetico-deductive (HD) approach, testing the validity of hypothesized theory in specific conditions, generally according to reductionist methodologies or designs, with the risk of over simplifying the initial complexity empirically perceived in its holistic view. However, in current food and nutrition research, new hypotheses are often elaborated from reductionist data obtained with the HD approach, and aggregated to form (ultra)reductionist theories, with no application of EI observations, limiting the applicability of these hypotheses in real life. This trend and the application of the EI method are illustrated as regards with the global health issue through the examples of food classifications/scoring, clinical studies, the definition of a sustainable diet, the "matrix effect"-related hypothesis, the concept of healthy core metabolism, and obesity prevention within the perspective of social sciences. To be efficient for producing food and nutritional data appropriable by the society, it finally appears that not only both approaches are necessary, starting with the EI method then the HD one, but also a back and forth between the two, this being not always realized, potentially leading to confusion and misunderstanding in society.