The staging of cancer based on its anatomical extent is intended to provide a precise yet concise description of tumor spread, within the context of multidisciplinary care and clinical trials; its use is also essential in epidemiological studies describing trends in cancer incidence or survival. Accurate staging poses challenges in the context of systematic population-based registration. These challenges, and the potential biases resulting from them, are illustrated by examples from a study of colorectal cancers diagnosed in 1990, conducted by six French registries. The problems posed by the evaluation of the components of the TNM classification (particularly classification shifts) are presented, along with the difficulties encountered in reconstructing stages. The complexity of correctly applying stages in observational settings explains the caution with which results must be interpreted and the need for standardization of stage recording by registries.