Text categorization is a process of assigning labels to documents according to the contents or topics of the documents. Traditionally text categorization is carried out by human experts. However, due to factors such as blurred category boundaries, background bias, and personal judgment, label inconsistency is often found in human classified collections, thus reducing their values in various applications. This article described an automatic process to detect such inconsistency based on the Agricultural Science Information Center (ASIC ) collection. In the article, important examples and results are presented. Potential benefits and applications are discussed.