Iron ore is a polycrytalline material created by complex natural process during geological period, wich give rise to intrinsic characteristics and varied industrial behavior. The vast majority of the Brazilian iron ores belong essentially to the hematitic type. Hematite can be classified as lobular, lamelar, granular micro-crystalline or martite. In the mineral industry, the characterion of iron ore and its agglomerates is traditionally developed by human operatorsform the observation of samples under the optical microscop, wich may suffer large variations. Thus, it is important to develop a procedure that allows the discrimation of the different hematite types and the measurement of characteristics suchs crystal size. The present thesis proposes a system for the automatic classification of hematite textural types, based of digital on processing and analysis of optical microscopy images, in bright field, linear and circular polarized light. Routines were developed for the acquisition, registration, recognition and morphological analysis of hematite crytals. The automatic segmentation of hematite crystals was based on calculating the spectral distance, in order to control the region expansion form the seeds. The results regarding the identification of the obtained cystals were very promising. Size and shape attributes were obtained and used as a training set for supervised classifiers, leading to the recognition of granular, lamelar and lobular hematite classes. Global sucess rates close to 98 percent were obtained concerning self-validation as well crossed validation.