Since Web 2.0 is the largest container for subjective expressions about different topics or issues expressed in all languages, the study of Sentiment Analysis has grown exponentially. In this work, we focus on Spanish polarity classification of hotel reviews and a new domain-dependent lexical resource (eSOLHotel) is presented. This new lexicon has been compiled following a corpus-based approach. We have carried out several experiments using an unsupervised approach for the polarity classification over the category of hotels from corpus SFU. The results obtained with the new lexicon eSOLHotel outperform the results with other general purpose lexicon.