An approach based on the word frequency is proposed in this paper to estimate the size of Web database. It obtains a random sample on a certain attribute by analyzing the attribute correlations among all the textual attributes in the query interface. The size of a Web database can be estimated by submitting probing queries which are generated by top-k frequent words to the query interface of a Web database. The experiments on several real-world databases have proved that this approach is effective and can achieve high accuracy in estimating the size of Web databases.