Crack propagation in heterogeneous media is of primary interest for engineering purposes, in order to predict the overall toughness and the probability of fracture from data on the microstructure. Probabilistic models for mode I crack propagation in two dimensions are presented. They are developed for brittle elastic materials with a random distribution of fracture energy. These models enable us to calculate in a closed form the probability of fracture involving crack nucleation and propagation that differ from the usual fracture statistics models based on the weakest link model. The use of the Griffith’s crack arrest criterion is applied to random function models for the distribution of the fracture energy and for various loading conditions resulting in stable or unstable crack propagation. From the models are deduced some statistical size effects.