Abstract: Traditional key word filtering technology meets people's common need, but the flexibility and effect is too limited to recognize or filter the transformed key words. Semantic elements were applied to net monitor and a new key word filter method was proposed. This method could recognize and filter the transformed key words effectively. The filter time was O(L) rather than general O(LN), where L was the length of text and N was the size of Keyword-base. It means this algorithm costs constant time even if N is hundreds of thousands or millions. It is very practical in net monitor and information filter.