This paper describes an unsupervised system for sentiment analysis is presented in Spanish. The system performs a complete fine grain analysis, where the most important characteristics or aspects of an opinion are identified in order to determine their sentiment associated. The unsupervised approach used allows to extract, identify and sentiment classify, from the analysis of opinions in Spanish in a specific domain, allowing to scale to another language and domain with great ease. For the validation of AspectSA, several experiments were carried out using corpus of opinions in the restaurant domain. The results obtained exceeded the majority of existing systems for the Spanish language.