This paper presents a new optimization procedure for Stirling engines based on neural network concepts. Based on modeling and experimental data an intelligent and fast method is proposed which finds the best values for different design variables. Design variables of Stirling engine are optimized using Multi-Layer Perceptron (MLP) neural networks. The optimization procedure is performed for three typical design variables for a given precision which has the capability to be extended for various types of engines and designs. Design variables are assumed to be phase angle, displacer stroke and working frequency of ST500 Stirling engine. Output variables which are subject to be optimized are the engine output power and efficiency. Maximizing a weighted average of these output variables is defined as total procedure objective. Usually, output variables are calculated using mathematical modeling based on structure, drive mechanism, dynamics and thermodynamic properties of the Stirling engine, The Multi-Layer Perceptron network is trained via 125 samples of design and output variables, generated by Nlog thermodynamic analysis code, in order to learn the relations between them. Then, the maximum point for the weighted average of the network outputs is determined with adequate precision. The proposed optimal point for the Stirling engine finally is substituted into the analysis code, and its outcome is compared with network results at this point. The proposed values are validated by analysis code. In order to check the validation of Nlog analysis code results for this scope of study, a test setup for ST500 engine is also established. Absolute pressure of working fluid and crank angle of the engine are measured instantly for three different charge pressures and compared with analysis results. (C) 2014 Elsevier Ltd. All rights reserved.