Experiments of Parvizi et al. on transverse fracture of cross-ply laminates showed that both energy (Griffith) and strength criteria are necessary conditions for fracture but neither one nor the other are sufficient. Thanks to the singularity at the tip of the notch, the incremental form of the Griffith criterion gives a lower bound of admissible crack lengths. On the contrary, the strength criterion leads to an upper bound. The consistency between these two conditions provides a general form of a criterion for crack nucleation. (C) 2001 Academie des sciences/Editions scientifiques et medicales Elsevier SAS.