Abstarct: Given that natural gas is a mixture of several elements, we need to know precisely the compositon of the elements in the gas to determine its thermodynamic properties. Therefore, in this research, the thermodynamic properties of natural gas are calculated using multi-laxyer perceptron neural networks method. In oreder to train neural network, a data set is needed first that in current study, the data required for training the neural networks are obtained from ISO 20765-1 standard which is the updated form of AGA8 equation of state and three sets of neural networks are trained by these data that first set of neural networks with inputs: gas compositions, temperature and pressure are able to accurately calculate nine thermodynamic properties, including compressibility factor, isochoric heat capacity, isobaric heat capacity, speed of sound, isentropic exponent, enthalpy, entropy, internal energy and joule-thomson coefficient. The second neural network set with inputs: molar fractions, pressure and enthalpy and third neural network set with inputs: molar fractions, pressure and entropy are able to calculate all thermodynamic properties of natural gas. Examples of application of trained neural networks include the accurate calculation of flow rate of natural gas and the calibration of natural gas flowmeters and the analysis of thermodynamic processes, such as precise calculation of temperature drop in joule-thomson valves and turbo expanders.