The physical integrity of a database can be violated by many different kinds of software and hardware failures. Well-known examples for these kinds of failures are application program interrupts (e.g. due to a division by 0), system crashes (e.g. due to a power failure), and disk crashes (e.g. due to some kind of mechanical concussion). To restore the database to a consistent state in these and other situations, appropriate database logging and recovery mechanisms must be provided and must be integrated into the Database Management System (D BMS). This paper presents the logging and recovery strategies of the Advanced Information Management Prototype (AIM-P), a DBMS prototype which has been designed and implemented in a research project at the IBM Heidelberg Scientific Center. AIM-P transaction management is heavily based on the use of so-called 'Transaction-Oriented Work Spaces' (short TWSs ). Each TWS can be seen as a private database partition (segment) where a running transaction can store its modified data before the end of the transaction. The TWS can be used, as we will show, for efficient transaction undo and redo processing in case of a failure. In the present paper, the TWS concept is introduced, its effects on AIM-P transaction management are explained, and its usage for database recovery is outlined.