Combustion chemistry models have been developed with inherent uncertainties in them. Whether a model that is developed using fundamental combustion experiments is capable to reproduce practical combustion processes within typical levels of measurement uncertainty is an open question. This paper quantifies the uncertainty of engine autoignition simulation using the uncertainties in the selected chemical kinetic model and then minimizes the model prediction uncertainty using various experiments. The method adopts a deterministic framework of uncertainty quantification, termed bound-to-bound data collaboration, and applies it to simulate the autoignition of n-pentane in a standard octane rating experiment. The results show that simulation of the end-gas autoignition using a comprehensively tested n-pentane model coupled with a two-zone engine combustion model yields an uncertainty substantially higher than that of engine experiment (as indicated by the cycle-to-cycle variation of the autoignition timing measurement). In-cylinder thermochemical conditions are found to be less important than the kinetic parameters in determining the model uncertainty. The large model uncertainty can be reduced by constraining the simulation with consistent experimental data and their measurement uncertainties, including those from fundamental experiments that measure ignition delays, species concentrations, flame speeds, and more significantly from autoignition experiments in well-calibrated engines. (C) 2020 The Combustion Institute. Published by Elsevier Inc. All rights reserved.