Around fifteen years before the chemical substance alumina could be decomposed in the laboratory, it was identified as a compound and predicted to contain a new element called ‘aluminium’. Using this episode from early nineteenth-century chemistry as a case study for the use of analogical reasoning in science, this paper examines how chemists relied on chemical classifications for the prediction of aluminium. I argue that chemists supplemented direct evidence of chemical decomposition with analogical inferences in order to evaluate the composition (...) of substances. Since they were established on the basis of relevant similarities in chemical properties, classifications were taken to reflect so-called ‘chemical analogies’ between substances. In combination with the knowledge that analogous properties indicated similarities in composition, this enabled chemists to infer the composition of experimentally indecomposable substances by analogy. The trust in such analogical inferences, even if they could not be confirmed through experiment, was justified by pragmatic considerations. Whereas the possibility of decomposing a substance depended on the available laboratory techniques, chemical analogies were thought to last independently of internal composition. It was therefore more practical to keep well-established classes of substances together rather than separate them on the basis of experimental results that might evolve as new techniques were developed. Besides highlighting the links between analogical reasoning and classification, this paper also illustrates the importance of analogy in the evaluation of elementary nature in the early nineteenth century. (shrink)