As one of the most promising energy storage mediums, Lithium-ion batteries (LIBs) have attracted extensive research interest. A major challenge associated with LIB application is thermal runaway, which can be triggered during overheating, overcharging, collision, and other abused conditions, leading to a direct threat to lives and properties. Here, we select Li-ion batteries with lithium cobalt oxide cathode and graphite anode (18650, Samsung), which are relatively simple and extensively investigated, to revisit thermal runaway. The experiment is conducted using an accelerating rate calorimeter (EV+ ARC, Thermal Hazard Technology) following the standard heat-wait-seek strategy, and is repeated for 9 similar new cells with 100% state of charge (SOC). Key features such as onset temperature of self-heating, the onset temperature of thermal runaway, maximum heat release rate, thermal runawaydelay time, and maximum temperature are measured experimentally. Although the cells are all brand new and have been initiated similarly, obvious cell-to-cell variability has been observed during thermal runaway. Major differences exhibit in the measured exothermic onset temperature, delay time, and mass losses; however, each cell nevertheless shows highly consistent activation energy processed from the heat release rate. Additionally, thermal runaway models are established in COMSOL to simulate the experimental conditions, with and without the effect of reactant consumption during the heating stage. It is shown that the widely used four-step thermal runway model cannot quantitatively capture the activation energy from the heat release rate. To improve the kinetic modeling and accommodate the cell-to-cell variability, statistical analysis is conducted to process the experimental results. Meanand standard deviation of the frequency factor and activation energy has been acquired to determine the lower and upper bound of the kinetic modeling using a one-step global chemistry. The measured and simulated thermal runaway delay time has reached a reasonable agreement, with the uncertainty of the kinetic model considered. The identified cell-to-cell variability is most likely due to manufacturing inconsistencies, which should not only be considered for cell-level safety evaluation but also the design and evaluation of battery modules and packs.