This paper presents a gas turbine design and off-design model in which the difficulties due to lack of knowledge about stage-by-stage performance are overcome by constructing artificial machine maps through appropriate scaling techniques applied to generalized maps taken from the literature and validating them with test measurement data from real plants. In particular, off-design performance is obtained through compressor map modifications according to variable inlet guide vane closure. The set of equations of the developed analytical model is solved by a commercial package, which provides great flexibility in the choice of independent variables of the overall system. The results obtained from this simulator are used for neural network training: problems associated with the construction and use of neural networks are discussed and their capability as a tool for predicting machine performance is analyzed.This paper was presented at the ECOS'01 Conference in Turkey, July 4-6, 2001 and revised