The methane number (MN) of natural gas is predicted using mathematical modeling and machine learning techniques which can be incorporated into a sensor. Natural gas quality is known to vary seasonally and regionally depending on the geological region. MN is defined by the gas composition and also related to the knocking resistance in a natural gas engine. This article presents the results of two different methods to predict MN, which are Multiple Regression (MR) and Support Vector Regression (SVR) that can be further specified by three kernel types: linear, polynomial, and Gaussian distributions. The analysis of the 4 methods - MR, linear SVR, polynomial SVR, and Gaussian SVR -shows that each predicts MN with a root mean square error of +/- 1.06, +/- 1.08, +/- 0.54, and +/- 0.20 respectively. About 37% of the predictions made by MR are under the +/- 0.5 error range. 40% of linear SVR, 52% of polynomial SVR, and 98% of Gaussian SVR predictions are within the +/- 0.5 error range. The SVR using Gaussian kernel outperforms the other three methods in accuracy. The results can enable the technology needed to develop an intelligent sensor that can estimate the MN of natural gas online and in real-time, economically and reliably, overall increasing fuel efficiency and emission performance.