We present a new model of fault nucleation and growth based on the Weibull theory, already widely used in fracture research engineering. We propose that, according to a birth-and-growth process, germs (nuclei) are born at random instants at random spatial locations and then grow with time. This leads to a satisfactory formulation of fault length distribution, different from classical statistical laws. Especially, this formulation reconciles previous analyses of fault datasets displaying power-law and/or exponential behaviors. The Weibull parameters can be statistically estimated in a simple way. We show that the model can be successfully fitted to natural data in Kenya and Ethiopia. In contrast to existing descriptive models developed for geological fault systems, such as fractal approaches, the Weibull theory allows to characterize the strength of the material, i.e. its resistance to deformation. Since this model is very general, we expect that it can be applied in many situations, and for simulations of geological fracture processes. The model is independent of deformation intensity and type and therefore allows a better constraint of the seismic risk in threatened regions.