As part of her seminal work on models and analogies in science, Hesse develops a general ‘network theory’ of meaning in which she posits that all language (both natural and scientific) is fundamentally metaphorical. Drawing on Wittgenstein’s theory of family resemblance, she claims that the application of any concept to a novel case is based on the tacit perception of complex similarities between the new case and cases already subsumed under the concept in question, and thus amounts to a metaphorical transfer of that concept to the novel case. In this chapter, it will be argued that Hesse’s resulting theory of scientific inference lapses into anti-realism, as Hesse regards the similarities based on which all descriptive concepts are applied not as objective features of the world, but as dependent on social or anthropological factors. In contrast, it will be shown that Wittgenstein’s original theory of meaning already makes room for a genuine epistemic access to objective resemblances in the world—one which is established through direct, practically embodied acquaintances of the cognitive agent with such resemblances. Wittgenstein’s overall theory thus proves to be more realist than Hesse appreciates in her own interpretation and application of it, and thus helps to redeem the realist program of Hesse’s approach.