Automatic detection of figurative constructions used in texts can be quite a tricky task. In order to make it possible for a program to detect such constructions in a text it is necessary to understand what principles underlie sarcastic and ironic utterances and thus can be used for their identification. The paper deals with the linguistic differentiation between irony and sarcasm based on the analysis of tweets. We discuss the linguistic peculiarities of eight corpora marked with #irony and #sarcasm hash tags.