It has been hypothesized that neoplastic progression develops as a consequence of an acquired genetic instability and the subsequent evolution of clonal populations with accumulated genetic errors1. Accordingly, human cancers and some premalignant lesions contain multiple genetic abnormalities not present in the normal tissues from which the neoplasms arose2,3. Barrett oesophagus (BE) is a premalignant condition which predisposes to oesophageal adenocarcinoma (EA) that can be biopsied prospectively over time because endoscopic surveillance is recommended for early detection of cancer4,5. In addition, oesophagectomy specimens frequently contain the premalignant epithelium from which the cancer arose6. Neoplastic progression in BE is associated with alterations in TP53 (also known as p53) and CDKN2A (also known as p16) and non-random losses of heterozygosity7,8,9,10,11 (LOH). Aneuploid or increased 4N populations occur in more than 90-95% of EAs, arise in premalignant epithelium and predict progression10,12,13. We have previously shown in small numbers of patients that disruption of TP53 and CDKN2A typically occurs before aneuploidy and cancer10,11,14,15. Here, we determine the evolutionary relationships of non-random LOH, TP53 and CDKN2A mutations, CDKN2A CpG-island methylation and ploidy during neoplastic progression. Diploid cell progenitors with somatic genetic or epigenetic abnormalities in TP53 and CDKN2A were capable of clonal expansion, spreading to large regions of oesophageal mucosa. The subsequent evolution of neoplastic progeny frequently involved bifurcations and LOH at 5q, 13q and 18q that occurred in no obligate order relative to each other, DNA-content aneuploidy or cancer. Our results indicate that clonal evolution is more complex than predicted by linear models.