Abstract A new combustion regime classification method is proposed for HCCI/PCCI combustion based on Lagrangian fluid particle tracking. The performance of the present method is demonstrated with the time trace of net heat conduction to local mixture which is one of the quantities best representing the phenomena. Local combustion regimes are classified based on magnitudes of the maximum and minimum heat conduction on each fluid particle throughout the combustion process into autoignition, flame propagation and flame enclosing. Direct numerical simulations (DNS) of turbulent HCCI/PCCI combustion are conducted with different initial temperature fluctuations. A detailed kinetic mechanism including 53 species and 325 elementary reactions is used for methane–air reaction and a reduced kinetic mechanism including 37 species and 61 reactions is adopted for n-heptane–air reaction. The present DNS results show that the combustion process is highly sensitive to initial temperature fluctuations. The proposed method is applied to DNS results and the characteristics of fluid element classified into each combustion regime are investigated. By adopting appropriate threshold for the maximum and minimum heat conduction to local mixture, the contributions of three combustion regimes to whole fuel consumption can be discussed. The autoignition regime is dominant for lower temperature fluctuation case. However, the flame propagation and flame enclosing regimes turn to be significant for higher temperature fluctuation case where the autoignition regime is less than 20% and only triggers combustion start. Such qualitative tendency is confirmed in two fuel species. However, quantitatively, the transition of dominant combustion regime from autoignition to flame propagation/flame enclosing takes place at lower temperature fluctuation for the n-heptane–air mixture case. The proposed classification method effectively characterizes the turbulent combustion process in HCCI/PCCI engines. This classification method can be extended to quantify reaction mechanisms by tracking representative quantities such as the mole fraction of key intermediates.