In-cylinder process of a direct injection (DI) compression ignition (CI) engine was studied by using the Rutgers high-speed spectral infrared (M) imaging system and the KIVA-II computer code. Comparison of the engine measurements with the computational prediction was attempted.In order to perform the instantaneous IR imaging, a Cummins 903 engine cylinder head was modified by installing an optical access in place of one of the intake valves, which required designing a new rocker-arm mechanism. The measurements obtained using the highspeed dual spectra IR imaging system were processed by the conventional two-color method which employed soot as the radiating target. The KIVA-II program was coded in order to match engine and operation conditions to those employed in the present measurements for achieving mutual consistency of the analysis.In the study, the thermochemical progress over a spray plume was measured by simultaneously obtaining IR digital images in two spectral bands at successive instants of time during the reaction period, which generated some new findings. In general the thermal image appeared strong after the premixed combustion stage. The results processed by the two-color method indicated that there were low temperature zones near the spray axis. This phenomena was explained to take place due to either the spray formation process involving the hollow spray injected into a narrow-height combustion volume or the unstable high-speed flows compounded by the latent heat of evaporation. The soot concentration distributions inferred from KL values were compared with the temperature distributions obtained from the spectral digital images. High soot concentration zones indicated by KL were found between the low temperature and high temperature zones. This is interpreted that zones at low temperatures were not expected to produce large amounts of soot because the reactions there were expected to be slow while those at high temperatures consumed the soot at high rates. Under the engine and operating conditions investigated in this study, the computational results failed to predict the measurement results.