How cells correct deviations from a mean cell size at mitosis remains uncertain. Classical cell-size homeostasis models are the sizer, timer, and adder 1. Sizers postulate that cells divide at some threshold size; timers, that cells grow for a set time; and adders, that cells add a constant volume before division. Here, we show that a size-based probabilistic model of cell-size control at the G2/M transition (P(Div)) can generate realistic cell-size homeostasis in silico. In fission yeast cells, Cyclin BCdc13 scales with size, and we propose that this increases the likelihood of mitotic entry, while molecular noise in its expression adds a probabilistic component to the model. Varying Cdc13 expression levels exogenously using a newly developed tetracycline inducible promoter shows that both the level and variability of its expression influence cell size at division. Our results demonstrate that as cells grow larger, their probability of dividing increases, and this is sufficient to generate cell-size homeostasis. Size-correlated Cdc13 expression forms part of the molecular circuitry of this system. Display omitted •A size-correlated division probability can generate cell-size homeostasis•Cyclin B concentration scales noisily with size in fission yeast•Cells with stochastically suprathreshold cyclin B are the ones that divide•A new tetracycline inducible promoter with linear dose response is developed Patterson et al. show that a probabilistic model of cell-size control can generate cell-size homeostasis in fission yeast. Cyclin B levels scale noisily with cell size, providing a molecular correlate for the phenomenological model. Using a new inducible promoter, they show that cyclin B and cell-size variability scale proportionally.