Our concept of cancer latency, the interval from when a cancer starts until it is diagnosed, has changed dramatically. A prior widely-used definition was the interval between an exposure to a cancer-causing substance and cancer diagnosis. However, this definition does not accurately reflect current knowledge of how most cancers develop assuming, mostly incorrectly, one exposure is the sole cause of a cancer, ignoring the possibility the cancer being considered would have developed anyway but that the exposure accelerated cancer development and eliding the randomness in when a cancer is diagnosed. We show, using chronic myeloid leukaemia as a model, that defining cancer latency is not as simple as it once seemed. It is difficult or impossible to know at which event or mutation to start to clock to measure cancer latency. It is equally difficult to know when to stop the clock given the stochastic nature of when cancers are diagnosed. Importantly, even in genetically-identical twins with the same driver mutation intervals to develop cancer vary substantially. And we discuss other confonders. Clearly we need a new definition of cancer latency or we need to abandon the concept of cancer latency in the modern era of cancer biology.