This paper describes a first test run of Sentitext, a sentiment analysis system under development. Unlike most existing systems, Sentitext is entirely based on linguistic knowledge and independent of any domain, using a wide coverage lexical database and lacking learning algorithms or classifiers, strictly speaking. Results on this first test, for which a collection of hotel review texts from Tripadvisor has been used, are extremely encouraging, given the high polarity hit rate. Keywords: Sentiment analysis, opinion mining, on-line user reviews. 1 Introduccion *