Summary form only given. Databases and their applications are topics of interest to both academia and industry. However, they have received little attention towards improving the knowledge about their associated faults and failures. This lack of knowledge is an impediment to the definition of adequate software testing techniques applicable in this domain and to the development of quality software. We discuss issues arising from SQL manipulation failures and present the results of an investigation aiming at understanding the relationship between faults and failures. The analysis of data mapping indicates that: i) there is a many-to-many mapping between faults and failures; ii) failure dimensions are dependent on fault type, faulty command, and the database itself; and iii) knowledge of the manipulation faults is crucial to programming and testing database applications.