In this paper we propose a new method for the identification of noisy regions in normalized iris images. Starting from a normalized and dimensionless iris image in the polar coordinate system, our goal consists in the classification of every pixel as "noise" or "not noise". This classification could be helpful in the posterior feature extraction or feature comparison stages regarding the construction of biometric iris signatures more robust to noise. We propose the extraction of 8 well known features for each pixel of the images followed by the classification through a neural network.