Extracting attribute names and values from textual product descriptions is important for many e-business applications such as user demand forecasting and product comparison and recommendation. The existing approaches first use supervised or semi-supervised classification techniques to extract attribute names and values, and then match them by analyzing their grammatical dependency. However, those methods have following limitations:(1) They require human intervention to label some attributes, values and the matching relationship between them; (2) The matching accuracy may be greatly affected by language habits, semantic logic, and the quality of corpus and candidates sets. To address these issues, this paper proposes an unsupervised approach for attribute name and value extraction and matching in Chinese textual merchandise descriptions. Taking advantage of search engine, it extracts the candidate set of attribute names with respect to a value by analyzing grammatical relation based on the principle of small probability event. A new algorithm for computing the matching probability between attribute names and values is also designed based on relative conditional deselect probability and Page Rank. The proposed approach can effectively extract attribute names and values from Chinese textual merchandise descriptions and match them without any human intervention, no matter whether the attribute name appears in the textual description or not. Finally, the performance of the proposed approach is evaluated on the textual descriptions of 4 types of merchandise using the search engine of Baidu. The experimental results show that the new approach for attribute name extraction can improve recall by 20%, compared with the approach of directly extracting attribute names from textual descriptions. Moreover, the new approach achieves considerably higher matching accuracy (above 30% if measured by the percentage of rank-1, above 0.3 if measured by MRR) than the existing techniques based on grammatical dependency analysis for non-quantization attributes.