Spoken or written texts are coherent sequences of sentences. Text comprehension is equivalent to the construction of multiple mental representations in working memory. It is based on an interaction between external text information and internal prior knowledge information stored in long-term memory. Mental representations include a text surface representation, a propositional representation, and a mental model. They are characterized by different forgetting rates. As speakers and authors omit information which can be easily completed by listeners and readers, text comprehension always includes inferences. Listening and reading comprehension use the same lexicon and the same syntax but qualitatively different text surface structures. Due to local and global coherence of texts, comprehension is also a process of mental coherence formation. Limitations of working memory require focused attention on the construction of topic-specific mental models which are carried along from sentence to sentence by a flow of consciousness. Speakers and authors can direct this process through topic information within the text surface.