A principal component analysis (PCA) scheme is developed for treatment of observations from the high spectral resolution Infrared Atmospheric Interferometer (IASI) spaceborne instrument. Compression and denoising of IASI observations are performed using this PCA. This preprocessing methodology also allows for a fast pattern recognition to obtain a first guess from a climatological data set. The performance of the compression, denoising, and multivariate first-guess retrieval are evaluated with a large diversified data set of radiosondes atmospheres including rare events. Overall, the instrumental noise in the overall observed IASI spectrum goes from 0.9 to 0.2 K after denoising. This analysis procedure will be used by Aires et al. <2002c> to retrieve simultaneously temperature, water vapor and ozone atmospheric profiles.