Lexical bundles are defined as a sequence of three or more words that indicate a statistical tendency to occur together in a register. This research focuses on identifying the frequency, structure, and function of lexical bundles in registers of regulations and letters. This is a corpus based research with inductive approach. The corpus of data in this research are 1,019,326 words that come from 5 different kinds of text, which are laws, presidential regulations, ministerial regulations, formal letters and business letters. The frequency limit used to identify lexical bundles in this study is 20 times in at least 5 different texts. The result shows that there are 345 lexical bundles with three to seven words sequences, such as sebagaimana dimaksud pada, oleh karena itu, and yang dimaksud dengan. From those five texts, laws has the most lexical bundles. Meanwhile, in the analysis of the structure of three words bundles from those five texts are dominated by the complete structural pattern (56.85%), such as verbal phrase (e.g. dilakukan dengan cara), whereas the incomplete structural pattern (43.15%) is dominated by verbal phrase + preposition with the word "yang" as its modifier (e.g. yang diatur dalam). Related to the function of lexical bundles, it was found that there were functions that were oriented to the regulatory and textual aspects, while there were no functions that were oriented towards the interaction between the writer and the reader.