One of important features in natural language processing is to find the root of a word. Stemming means to remove prefixes, suffixes, and infixes for finding the root of the word. Its aims are about to information retrieval, exploring text, machine for translation, and word look up based on its root. Stemming increases document retrieval by 10-50% in most of international languages, it also compresses the size of web-based table indexes documents up to 50%. In this paper, by analyzing stemming approaches, using structural methods, and deterministic finite automaton machine, applying 274 existing prefixes in language (linkage), a stemming system for Azerbaijani language is generated. Experimental result demonstrates that the proposed algorithm performs more than 97% accuracy.