Researchers often attempt to categorize tumors into more homogeneous subtypes to better predict prognosis or understand pathogenic mechanisms. In clinical research, typically the focus is on prognosis: the tumor subtypes are intended to be associated with specific responses to treatment and/or different clinical outcomes. In aetiological research, the focus is on identifying distinct pathogenic mechanisms, which may involve different risk factors. We used directed acyclic graphs to present a framework for considering potential biases arising in aetiological research of tumor subtypes, when there is incomplete correspondence between the identified subtypes and the underlying pathogenic mechanisms. We identified two main scenarios: (1) weak effect, when the tumor subtypes are identified through combinations of characteristics and some of these characteristics are affected by factors that are unrelated with the underlying pathogenic mechanisms; and (2) lack of causality, when the set of characteristics corresponds with a mechanism that is actually not a cause of the tumor of interest. Examples of the magnitude of bias that can be introduced in these situations are provided. Although categorization of tumors into homogenous subtypes may have important implications for aetiological research and identification of risk factors, the characteristics used to classify tumors into subtypes should be as close as possible to the actual pathogenic mechanisms to avoid interpretative biases. Whenever our knowledge of these mechanisms is limited, research into risk factors for tumor subtypes should first aim to causally link the characteristics to the pathogenic mechanisms.