We describe a supervised system that uses optimized Condition Random Fields and lexical features to predict the sentiment of a tweet. The system was submitted to the English version of all subtasks in SemEval-2017 Task 4.