The stringent exhaust emission regulations of Environmental Protection Agency has forced engine manufacturers to develop alternative combustion systems and clean alternative fuels. Lean-burn natural gas engines are attracting increasing attention as a desirable alternative, especially for stationary applications. In-cylinder pollution formation in these engines is significantly lower than that of diesel engines. However, the thermal efficiency and power density of lean-burn natural gas engines must be improved to be considered as an alternative to diesel engines. The higher octane number of methane, which is the dominant component of natural gas, allows the engines to be turbocharged and have higher compression ratio than conventional spark ignited engines. However, seasonal and regional variation in natural gas composition, which cause natural gas to have lower octane number, is the biggest constraint for the most efficient design of these engines. Under these conditions, building a competitive natural gas engine is not possible with conventional methods and requires costly engine tests. One alternative to costly engine testing is analytical engine modeling. However, conventional engine modeling tools do not properly include end-gas autoignition and multi-dimensional engine modeling codes coupled with detailed chemical kinetic codes such as KIVA-CHEMKIN combination may not be preferred by engine designers. This dissertation describes a new autoignition model that does not require extensive computational resources (2-dimensional computations from intake valve closing to exhaust valve opening take generally less than 5 minutes with a 333 MHZ personal computer), is easily portable to various computational environments, is easy to use and the accuracy is compatible with the accuracy of other subroutines of the models. It also considers variation of natural gas composition due to propane addition. Computation results show that the knock occurrence crank angle can be predicted within 2 degrees CA when the model is coupled to a Zero-Dimensional engine model, which was also developed for the present work. The results with the model incorporated into a Multi-Dimensional model (KIVA) are also promising. KIVA was able to predict if the engine was going to knock or not and also gave trends in the knock intensity.