In vivo, tissue-level, mechanical thresholds for axonal injury were determined by comparing morphological injury and electrophysiological impairment to estimated tissue strain in an in vivo model of axonal injury. Axonal injury was produced by dynamically stretching the right optic nerve of an adult male guinea pig to one of seven levels of ocular displacement (Nlevel = 10; Ntotal = 70). Morphological injury was detected with neurofilament immunohistochemical staining (NF68, SMI32). Simultaneously, functional injury was determined by the magnitude of the latency shift of the N35 peak of the visual evoked potentials (VEPs) recorded before and after stretch. A companion set of in situ experiments (Nlevel = 5) was used to determine the empirical relationship between the applied ocular displacement and the magnitude of optic nerve stretch. Logistic regression analysis, combined with sensitivity and specificity measures and receiver operating characteristic (ROC) curves were used to predict strain thresholds for axonal injury. From this analysis, we determined three Lagrangian strain-based thresholds for morphological damage to white matter. The liberal threshold, intended to minimize the detection of false positives, was a strain of 0.34, and the conservative threshold strain that minimized the false negative rate was 0.14. The optimal threshold strain criterion that balanced the specificity and sensitivity measures was 0.21. Similar comparisons for electrophysiological impairment produced liberal, conservative, and optimal strain thresholds of 0.28, 0.13, and 0.18, respectively. With these threshold data, it is now possible to predict more accurately the conditions that cause axonal injury in human white matter.