In this work a neural network methodology is presented, aimed to retrieve atmospheric parameters of meteorological interest such as temperature, water vapour and ozone profiles from upwelling high resolution infrared sensor spectra. Neural network approach has been developed on basis of the specification of the Infrared Atmospheric Sounding Interferometer (IASI), which is planned to be flown on the first European Meteorological Operational Satellite Metop in 2005. The performance of the neural network based inversion methodology has been evaluated by considering a suitable set of inversion exercises in which test cases are retrieved. The error analysis shows that temperature may be retrieved in the troposphere within 1-1.5 K accuracy that is very close to the IASI mission objective of 1 K in 1 km layers. Quite interesting results have been also obtained for water vapor and ozone.