The evidence that links classical protein-coding proto-oncogenes and tumor suppressors, such as MYC, RAS, P53, and RB, to carcinogenesis is indisputable. Multiple lines of proof show how random somatic genomic alteration of such genes (e.g. mutation, deletion or amplification), followed by selection and clonal expansion, forms the main molecular basis of tumor development. Many important cancer genes were discovered using low-throughput approaches in the pre-genomic era, and this knowledge is today solidified and expanded upon by modern genome-scale methodologies. In several recent studies, non-coding RNAs (ncRNAs), such as microRNAs and long non-coding RNAs (lncRNAs), have been shown to contribute to tumor development. However, in comparison with coding cancer genes, the genomic (DNA-level) evidence is sparse for ncRNAs. The coding proto-oncogenes and tumor suppressors that we know of today are major molecular hubs in both normal and malignant cells. The search for non-coding RNAs with tumor driver or suppressor roles therefore holds the additional promise of pinpointing important, biologically active, ncRNAs in a vast and largely uncharacterized non-coding transcriptome. Here, we assess the available DNA-level data that links non-coding genes to tumor development. We further consider historical, methodological and biological aspects, and discuss future prospects of ncRNAs in cancer.