A new species reduction method called the Simulation Error Minimization Connectivity Method (SEM-CM) was developed. According to the SEM-CM algorithm, a mechanism building procedure is started from the important species. Strongly connected sets of species, identified on the basis of the normalized Jacobian, are added and several consistent mechanisms are produced. The combustion model is simulated with each of these mechanisms and the mechanism causing the smallest error (i.e. deviation from the model that uses the full mechanism), considering the important species only, is selected. Then, in several steps other strongly connected sets of species are added, the size of the mechanism is gradually increased and the procedure is terminated when the error becomes smaller than the required threshold. A new method for the elimination of redundant reactions is also presented, which is called the Principal Component Analysis of Matrix F with Simulation Error Minimization (SEM-PCAF). According to this method, several reduced mechanisms are produced by using various PCAF thresholds. The reduced mechanism having the least CPU time requirement among the ones having almost the smallest error is selected. Application of SEM-CM and SEM-PCAF together provides a very efficient way to eliminate redundant species and reactions from large mechanisms. The suggested approach was tested on a mechanism containing 6874 irreversible reactions of 345 species that describes methane partial oxidation to high conversion. The aim is to accurately reproduce the concentration–time profiles of 12 major species with less than 5% error at the conditions of an industrial application. The reduced mechanism consists of 246 reactions of 47 species and its simulation is 116 times faster than using the full mechanism. The SEM-CM was found to be more effective than the classic Connectivity Method, and also than the DRG, two-stage DRG, DRGASA, basic DRGEP and extended DRGEP methods.