Almost all cancers are preceded by a prolonged period of clinical latency during which a combination of cellular events helps move carcinogen-exposed cells towards a malignant phenotype. Hitherto, investigating the fate of premalignant cells in vivo remained strongly hampered by the fact that these cells are usually indistinguishable from their normal counterparts. Here, for the first time, we have designed a strategy able to reconstitute the replicative history of the bona fide premalignant clone in an animal model, the sheep experimentally infected with the lymphotropic bovine leukemia virus. We have shown that premalignant clones are early and clearly distinguished from other virus-exposed cells on the basis of their degree of clonal expansion and genetic instability. Detectable as early as 0.5 month after the beginning of virus exposure, premalignant cells displayed a two-step pattern of extensive clonal expansion together with a mutation load approximately 6 times higher than that of other virus-exposed cells that remained untransformed during the life span of investigated animals. There was no fixation of somatic mutations over time, suggesting that they regularly lead to cellular death, partly contributing to maintain a normal lymphocyte count during the prolonged premalignant stage. This equilibrium was finally broken after a period of 18.5 to 60 months of clinical latency, when a dramatic decrease in the genetic instability of premalignant cells coincided with a rapid increase in lymphocyte count and lymphoma onset.