Leaf margin analysis (LMA) is a widely used method that applies present-day linear correlation between the proportion of woody dicotyledonous species with untoothed leaves (P) and mean annual temperature (MAT) to estimate paleotemperatures from fossil leaf floras. Previous works demonstrate that LMA shows regional constraints and to date, no equation has been modeled directly from Chinese forests. Fifty humid to mesic Chinese forests were chosen to understand the relationship between percentage of untoothed leaf species and MAT in China. Consistent with previous studies, the Chinese data indicate that P shows a strong linear correlation with MAT, but the actual relationship is a little different from those recognized from other regions. Among the several currently used LMA equations, the one resulting from North and Central American and Japanese data, rather than the widely used East Asian LMA equation, yields the closest values to the actual MATs of the Chinese samples (mean absolute error = 1.9 °C). A new equation derived from the Chinese forests is therefore developed, where MAT = 1.038 + 27.6 × P. This study not only demonstrates the similarity of the relationship between P and MAT in the Northern Hemisphere, but also improves the reliability of LMA for paleoclimate reconstructions of Chinese paleofloras.