Abstract. The recent rise in popularity of Twitter and its open API provides developers the opportunity to extract amounts of data which can be a thesaurus of information. This opportunity led to the development of an open source and open API system called Flu track (http://flutrack.org) that monitors influenza epidemics, based on geo-located self-reports on Twitter. In particular, we detect words such as sore throat, cough, fever etc. Moreover, we detect the aggravation of a patient’s clinical condition when a user posts a second flu related tweet that contains words indicating further symptoms such as: worse, deteriorating. Finally, we present flu-positives with real time anonymous visualizations using maps (mapping), which might be helpful for authorities and sensitive populations to plan upcoming events or activities. In order to further aid the surveillance of the spreading of the disease, a classification experiment has been conducted for automatically identifying Tweets that describe cases with acute and more critical symptoms from those referring to milder cases. We found that making use of mereley very small n-gram keyword lexica, the automatic identification of critical cases reaches an accuracy of 92%.