The in vivo erythropoiesis, which is the generation of mature red blood cells in the bone marrow of whole organisms, has been described by a variety of mathematical models in the past decades. However, the in vitro erythropoiesis, which produces red blood cells in cultures, has received much less attention from the modelling community. In this paper, we propose the first mathematical model of in vitro erythropoiesis. We start by formulating different models and select the best one at fitting experimental data of in vitro erythropoietic differentiation. It is based on a set of linear ODE, describing 3 hypothetical populations of cells at different stages of differentiation. We then compute confidence intervals for all of its parameters estimates, and conclude that our model is fully identifiable. Finally, we use this model to compute the effect of a chemical drug called Rapamycin, which affects all states of differentiation in the culture, and relate these effects to specific parameter variations. We provide the first model for the kinetics of in vitro cellular differentiation which is proven to be identifiable. It will serve as a basis for a model which will better account for the variability which is inherent to experimental protocol used for the model calibration.vitro context, i.e. the process that takes place in cells grown in culture, is much simpler to characterize experimentally. Yet, to our knowledge, no modeling study has focused on it so far. Since the in vitro differentiation is an experimental tool of choice for the study of cellular decision-making (17,18,19) , we propose to develop a model for the dynamics of the in vitro erythropoiesis.Moreover, the current models of erythropoiesis suffer from one major drawback: the weakness of their parameterization, which can fall within three categories.A vast majority of the existing models of erythropoiesis are based on experimental parameter values from the litterature. In some cases these values are used in other contexts that those in which they were obtained (typically, in other species (12) ).In other cases, the parameter values of a model are chosen arbitrarily to reproduce a qualitative behaviour. Apart from this qualitative fit, such approaches do not provide any information regarding the validity of the values (16) .Finally, when the parameters of a model are estimated to reproduce a dataset, the precision of this estimation is seldom investigated (20) . By this, we mean that depending on the algorithmic details of the estimation, it is possible that several values of the parameters might render the same fit to the data. In this case the model is said to be unidentifiable.A model is said to be identifiable if and only if it is possible to infer a unique value for each of its parameter by comparing its output to experimental data. Otherwise it is unidentifiable. A model can be non-identifiable for several reasons (21,22) .Structural identifiability is related to the structure of the model, and the observed variables. A model is structurally unidentifi...