Arabic is a highly inflected language, and therefore the processes of stemming and root extracting represent a challenge to researches. A new method is presented for extracting Arabic text stem, and lemma. Stemming sometimes affects the semantic of a word, where as lemma preserve the meaning of a word. The approach is based on pattern extraction. It uses a special encoding based on dividing letters into original and non-original letters. Codes are automatically generated for each pattern and then match against input text to extract root, pattern, and lemma of a word. A comparison with other methods reveals a promising result with accuracy up to 96%.