PURPOSE. To describe and evaluate a semiquantitative optic nerve grading scheme for assessing axonal loss in endothelin (ET)-1-induced chronic optic neuropathy. METHODS. Optic nerve cross-sections from both eyes of 39 Brown Norway rats unilaterally treated with various concentrations of ET-1 or physiological saline solution via a surgically implanted osmotic minipump were processed for light and transmission electron microscopy (TEM). The optic nerve damage grade, between 0 (no damage) and 10 (total damage), was based on the number of zones of approximately equal damage and the mean percentage of damage within each zone. Grading was performed under light microscopy by three observers and compared with axonal survival determined with TEM using two quantification methods: the sampling method, in which ∼10% of the section was counted, and the full-count method, in which the whole section was counted (n = 12). Axonal survival was expressed as a ratio of axon counts in the experimental to control eye. Before these comparisons, the interand intraobserver agreement rates were determined in another group of 85 and 12 ET-1-treated animals, respectively. RESULTS. The interobserver κ was 0.66 (95% confidence interval [CI]: 0.58-0.74) for all eyes and 0.55 (95% CI: 0.43- 0.67) for the experimental eyes only. The intraobserver κ was 0.80, 0.81, and 0.80 for all 24 eyes and 0.60, 0.64, and 0.71 for experimental eyes only. The correlation between damage grade in the experimental eye and axonal survival using the TEM sampling method (Spearman's ρ = -0.677 for all animals and -0.827 for the subset of animals with full counts only) was lower than that with the full-count method (Spearman's ρ = -0.926). When axonal survival was less than 0.7, the sampling method always underestimated the extent of damage. CONCLUSIONS. The grading scheme had good inter- and intraobserver agreement, and high correlation with the TEM methods. It is a practical and time-saving method, requiring less than 1 minute per nerve and is an alternative to sampling methods that can yield significant errors.