Lumping of isomers is an effective approach to reduce large chemical kinetic mechanisms containing a substantial number of chemical species. However, a simple lumping of all isomers with identical chemical compositions leads normally to large prediction uncertainties of the lumped model, by neglecting the chemical functionalities of isomers in terms of, for instance, their oxygenated groups and radical sites, which affect their formation and consumption pathways significantly. Thus, the definition group of lumped species is crucial, which is, however, typically done manually in the model reduction by trial and error. In the combustion community, large kinetic mechanisms are mostly developed based on deriving chemical reactions and their involved species according to the so-called reaction class, which produces a large number of molecules with identical compositions but different structures and chemical functionalities. Thus, in this study, an automatic chemical lumping approach is proposed to efficiently reduce the kinetic mechanisms of large hydrocarbon fuels by defining the group of lumped isomers according to their reaction classes and chemical functionalities automatically. For the species produced by the low-temperature reaction classes, such as hydroperoxyalkyl radicals, peroxy hydroperoxide radicals, and cyclic ethers, their isomers are lumped together, if their formation reactions have similar chemical characteristics with respect to the ring sizes of transition states. Reaction rate constants of lumped reactions are evaluated based on numerical results of the detailed mechanism. By using the derived method, a mechanism of n-decane with 1692 species is reduced to 137 species and a skeletal mechanism of n-heptane, iso-octane, toluene, and ethanol mixture with 489 species is reduced to 298 species successfully, in conjunction with the directed relation graph with error propagation (DRGEP) method. Agreement is observed between lumped and original mechanisms, which demonstrates the validity and effectiveness of the present approach.