To reduce the Web log data scale and discover more recommendable access patterns from data preprocessed, with knowledge based on amount of information, the concept of quantify value of importance of every property in relation to property set was proposed, and used the idea of LRU page replacement algorithm in the operating system, a new data preprocessing method based on importance of property was proposed. The experiments show that the method could delete Web log records which were caused by user short-behavior and have not mining value, and filter out the noise data. Accordingly it can reduce the complexity of log mining effectively.