Toxic keratopathy (TK) involves complex clinical manifestations and is difficult to differentiate from other ocular surface diseases by conventional slit-lamp examination. The challenge faced by clinicians in confidently diagnosing TK cannot be underestimated. This study aimed to explore the microstructural characteristics and diagnostic parameters by in vivo confocal microscopy (IVCM) in TK.In this prospective, cross-sectional, comparative study, slit-lamp and IVCM examinations were performed on 20 normal eyes and 54 eyes with TK. Based on slit-lamp imaging, TK subjects were divided into four groups: superficial punctate keratitis (n = 10 eyes), pseudodendritic keratitis (n = 14 eyes), ulcerative keratitis (UK; n = 16 eyes), and ring keratitis (RK; n = 14 eyes). The microstructural characteristics of TK were described according to the following IVCM parameters: basal cell (BC) density, dendritiform cell (DC) density, DC size, corneal nerve fiber (CNF) length, nerve tortuosity, and keratocyte reflectivity. A receiver operating characteristic (ROC) curve model was also formulated to compare the predictive power of BC density, DC density, and CNF length.TK eyes showed significantly higher values for DC density (45.8 cells/mm2; range, 25.0-100.0) compared with normal eyes (24.0 cells/mm2; range, 20.8-32.3; P = 0.013; DC size (111.0 µm2; range, 92.0-137.8) compared with normal eyes (63.7 µm2; range, 47.7-70.3; P < 0.001); nerve tortuosity (0.08; range, 0.05-0.09) compared with normal eyes (0.04; range, 0.02-0.04; P < 0.001); and keratocyte reflectivity. BC density and CNF length values were found to be significantly less than those for normal controls (both P < 0.001). In all subgroups, CNF length appeared to be significantly lower than that of controls (all P < 0.001), and DC density was only statistically significantly higher in the UK (P = 0.003) and RK (P < 0.001) groups. Corneal fluorescein staining had no correlation with the analyzed IVCM parameters (all P ˃ 0.05). However, the increase in DC density and DC size showed negative correlations to CNF length (density: r = -0.325, P < 0.005; size: r = -0.493, P < 0.005), as well as positive correlations to duration and frequency of topical eye drops and DC size (density: r = 0.361, P < 0.05; size: r = 0.581, P < 0.05). A ROC curve showed that CNF length had the strongest predictive power, with the estimated area under the curve being 0.992 ± 0.008.Lower BC density and CNF length, greater DC density and DC size, and greater keratocyte reflectivity were the microstructural characteristics of TK. The role of subbasal nerve, inflammatory response, and limbal stem cells in the progression of TK and the appropriate treatment of different TK stages are future research directions.The evaluation of basal cells, subbasal nerve, and dendritiform cells is helpful to our understanding of the pathological process of TK.