Since F T Lewis' pioneering work in the 1920s, a linear correlation between the average in-plane area of domains in a two-dimensional (2D) cellular structure and the number of neighbors of the domains has been empirically proposed, with many supporting and dissenting findings in the ensuing decades. Revisiting Lewis' original experiment, we take a larger set of more detailed data on the cells in the epidermal layer of Cucumis, and analyze the data in the light of recent results on size–topology correlations. We find that the correlation between the number-of-neighbor distribution (topology) and the area distribution is altered over that of many other 2D cellular systems (such as foams or disc packings), and that the systematic deviation can be explained by the anisotropic shape of the Cucumis cells. We develop a novel theory of size–topology correlation taking into account the characteristic aspect ratio of the cells within the framework of a granocentric model, and show that both Lewis' and our experimental data is consistent with the theory. In contrast to the granocentric model for isotropic domains, the new theory results in an approximately linear correlation consistent with Lewis' law. These statistical effects can be understood from the increased number of configurations available to a plane-filling domain system with non-isotropic elements, for the first time providing a firm explanation of why Lewis' law is valid in some systems and fails in others.