There are prevailing concerns with the critical dimensions when conventional theories break down. Here we find that the Griffith criterion remains valid for cracks down to 10 nm but overestimates the strength of shorter cracks. We observe the preferred crack extension along the zigzag edge in graphene, and explain this phenomenon by local strength-based failure rather than energy-based Griffith criterion. These results provide a mechanistic basis for reliable applications of graphene in miniaturized devices and nanocomposites. KEYWORDS: Griffith criterion, graphene, fracture, edge energy, molecular dynamics M ost mechanical theories used in engineering practice are developed for relatively large structures. Their applicability in nanoscale systems is however uncertain. The continuum theories usually work well for the collective behavior of a large ensemble of basic building blocks like atoms or molecules but are not guaranteed to hold when they are applied to systems composed of a limited number of basic units. Indeed, deviation of certain properties in small systems from their bulk counterparts is foreseen by Feynman. 1 The ensuing question is whether or not there is a critical size for a theory, below which this theory fails to adequately capture the physical behavior of a miniaturized system. For engineering applications, knowing such critical size is of paramount significance for the design of small structures that are mechanically reliable. The Griffith criterion of brittle fracture is one of the most fundamental theories for characterizing the mechanical behavior of defective materials and has been widely used in engineering design. 2,3 In this work, we explore the physical limit of the Griffith criterion, that is, when it fails to predict the fracture strength of materials with small nanosized cracks. The challenge to address such question of size limit lies in the difficulty to bridge the studies of microscopic and macroscopic systems by either experiments or atomistic simulations. On one hand, while progress has been recently made in the nanomechanical testing of materials, 4−9 it is still difficult to conduct a series of controlled experiments by systematically reducing the sample size while retaining the consistent microstructural features. On the other hand, molecular dynamics (MD) simulations can effectively explore brittle fracture at the atomic scale, 10−14 but full atomistic simulations are yet too computationally expensive to simulate three-dimensional systems with feature size larger than 100 nm.To understand the applicability of the continuum mechanics theory to a discrete atomistic system, it is necessary to probe the mechanical behavior of sufficiently large samples with atomic resolution. To this end, graphene is an ideal model material to investigate the critical size when the Griffith criterion of brittle fracture breaks down. This is because the monolayer graphene is only one atom thick 15 such that the inplane dimensions can be taken to be adequately large (e.g., up to microme...