The modelling of natural gas combustion with detailed as well as reduced kinetic mechanisms has been subject to many publications already. Most of the mechanisms are validated at higher temperatures and ambient pressures. Only recently - in parallel to a gain of interest in the homogeneous charge compression ignition (HCCI) engine concept - there are more efforts trying to reproduce the low temperature chemistry, and in particular autoignition of natural gas under engine-like conditions. This work is part of a project on stationary natural gas engines equipped with unscavanged prechambers. It has been demonstrated that it is possible to reduce the emissions below the Swiss limits both with natural gas and biogas with the spark ignition prechamber, while keeping high efficiencies of around 38 %. We are now aiming at converting the prechamber from spark ignition to autoignition leading to longer maintenance intervals in the case of stationary natural gas engines and to a probable cleaner combustion - HCCI like combustion inside the prechamber and faster combustion in the main chamber due to an earlier arrival of the flame front - compared to prechamber spark ignition. In order to be able to model the fluid dynamics and the chemistry inside the cylinder, a good chemical model is one of the most important aspects. To obtain an appropriate mechanism - by validating and/or optimising existing mechanisms - experiments have been conducted in the rapid compression machine facility at the University of Lille with different mixtures of methane, ethane and propane. The relative air-to-fuel ratio λ varied from 1 to 1.6, the temperature and pressure at the end of compression varied from 850 to 925 K and 13 to 21 bar respectively. The compression ratio was constantly kept at 9.2 for all experiments and the final temperature and pressure were defined by varying the charge and the composition of the inert gases (Ar/N2). The compression time was between 50 and 60 ms for all experiments. For modelling the ignition delay a zero-dimensional model assuming homogeneity was used, based on the measured pressure and calculated core temperature at the end of compression of each experiment. The chemical solver package used for the simulations is CHEMKIN. Different mechanisms (PRF, GRI3.0, Konnov, …) have been tested. Most of the mechanisms are mainly validated at higher temperature ranges and therefore their applicability to ignition delay prediction in our conditions is questionable. Several attempts to adjust the GRI mechanism (version 1.2) to the lower temperature regime exist, whereof we chose the UBCMech and Petersen mechanism (GRI+RAMEC) to test on our simulations. Both mechanisms initially were developed based on pure methane mixtures while the UBCMech recently was extended to model natural gas as well. The experimentally obtained ignition delays vary from 40 ms at high pressures and temperatures to more than 300 ms for the lower regime of the experimental space. For the ignition delays up to 60 ms the best fit was obtained by a reduced version of PRF accounting only for the C1 to C3 compounds. Konnov’s mechanism and UBCMech are under predicting the experimental ignition delays while the GRI3.0 mechanism (with and without RAMEC sub module) over predicts the ignition delay for the natural gas mixtures. It has to be mentioned that the GRI3.0+RAMEC mechanism as well as the initial mechanism of Petersen (GRI1.2+RAMEC) on the other hand under predict the experimental ignition delay for pure methane mixtures. This indicates that the low temperature chemistry for methane that has been added in the RAMEC sub module is enhancing ignition too strong. Replacing part of methane in pure methane mixtures by equivalent amounts of ethane and propane - know as ignition promoters - lead to longer ignition delays in the simulations. But as the GRI3.0+RAMEC mechanism (without nitrogen chemistry) is the smallest of all tested mechanisms considering number of species and reactions, it still is an interesting option for coupling it to a computational fluid dynamics code as we are planning to do. We though performed a sensitivity analysis to obtain a set of the most dominating reactions and used a multi-objective optimisation algorithm to optimize the reaction rates within their limits of uncertainty. The results show a good fit for the short ignition delays of natural gas mixtures but still an under prediction of the methane delays. For the longer ignition delays the assumption of a constant core temperature during the post compression phase becomes questionable and the under prediction of all mechanisms of the experimental delays is an obvious consequence. A heat loss model predicting the temperature evolution has to be implemented in the simulations to properly simulate the ignition delays.