Traditionally, SQL is the main interface to access data from relational databases. However, it is difficult for inexperienced end users to learn the complicated syntax of SQL. Enabling keyword-based information retrieval over relational databases will allow users to acquire information from databases without any knowledge of SQL and the underlying database schema, just like the way of common search engines. This paper describes the design and implementation of SEEKER, a system supporting keyword-based information retrieval over relational databases. While there have been some existing systems that support searching text attributes in relational databases, SEEKER can also search database metadata and numeric attributes. Moreover, SEEKER employs an improved ranking function and supports Top-k queries. Experimental results show that SEEKER can achieve good retrieval performance.