Single-cell transcriptomics has significantly advanced our ability to uncover the cellular heterogeneity of tumors. A key challenge in single-cell transcriptomics is identifying cancer cells and, in particular, distinguishing them from non-malignant cells of the same cell lineage. Focusing on features that can be measured by single-cell transcriptomics, this review explores the molecular aberrations of cancer cells and their observable readouts at the RNA level. Identification of bona fide cancer cells typically relies on three main features, alone or in combination: i) expression of cell-of-origin marker genes; ii) inter-patient tumor heterogeneity; iii) inferred copy-number alterations. Depending on the cancer type, however, alternative or additional features may be necessary for accurate classification, such as single-nucleotide mutations, gene fusions, increased cell proliferation, and altered activation of signaling pathways. We summarize computational approaches commonly applied in single-cell analysis of tumoral samples, as well as less explored features that may aid the identification of malignant cells.