The use of thermal infrared (TIR) remote sensing, between 8 and 14 μm, is an important method to monitor vegetation water stress status. To comprehend the spectral characteristics of natural leaves, understanding the relationship between TIR reflectance and leaf water content is helpful. In this study, we selected Magnolia cylindrica E.H.Wilson, Canna indica Rosc. and Gardenia jasminoides J. Ellis leaves to isolate cuticles with varying thicknesses. Subsequently, we constructed a leaf thermal infrared model (Leaf-TIR model) to examine the impact of cuticle thickness and inside wet epidermal cell wall on the TIR reflectance of natural leaves. Our findings indicate that the cuticle thicknesses of the three selected leaves are 12.2, 5.5, and 1.1 μm, respectively. Moreover, as cuticle thickness decreases, the similarity between the isolated cuticle and the leaf reflectance behaviors also decreases because a thinner cuticle has lower absorptance and a weaker effect on the reflection characteristics of natural leaves. Moreover, we observed that as the water content in the cell wall decreases, the reflectance of the leaves significantly increases when the cuticle thickness becomes 5.5 or 1.1 μm. This outcome is due to the increasing difference in the refractive indices of the cuticle and cell wall with declining water content. Our findings suggest that the Leaf-TIR model can quantitatively determine the relationship between TIR reflectance and water content in the cell wall. This information is crucial in interpreting TIR spectral behaviors of natural leaves and provides an important theoretical basis for future research.