Introduction: Many theoretical treatments assume (often implicitly) that delusions ought to be taxonomised by the content of aberrant beliefs. A theoretically sound, and comparatively under-explored, alternative would split and combine delusions according to their underlying cognitive aetiology. Methods: We give a theoretical review of several cases, focusing on monothematic delusions of misidentification and on somatoparaphrenia. Results: We show that a purely content-based taxonomy is empirically problematic. It does not allow for projectability of discoveries across all members of delusions so delineated, and lumps together delusions that ought to be separated. We demonstrate that an aetiological approach is defensible, and further that insofar as content-based approaches are plausible, it is only to the extent that they implicitly link content to aetiology. Conclusions: We recommend a more explicit focus on cognitive aetiology as the grounds for delusion taxonomy, even when that would undermine traditional content-based boundaries. We also highlight the iterative and complex nature of evidence about aetiologically grounded taxonomies.