This paper critiques the ANSI SQL isolation levels, which define isolation based on three phenomena: Dirty Read, Non-Repeatable Read, and Phantom. The authors argue that these definitions are ambiguous and fail to accurately characterize popular isolation levels, including those used in commercial databases. They propose a broader interpretation of these phenomena and introduce new ones to better define isolation types. An important multiversion isolation level, Snapshot Isolation, is defined, which avoids the ANSI SQL phenomena but is not serializable. Snapshot Isolation provides a reduced-isolation level between READ COMMITTED and REPEATABLE READ and is shown to be more powerful than the ANSI SQL isolation levels. The paper also discusses other isolation levels, including Cursor Stability and Snapshot Isolation, and their relationships. It concludes that the ANSI SQL isolation levels are incomplete and that new phenomena are needed to properly define isolation levels. The paper also highlights the importance of considering the broader implications of isolation levels and the need for a more comprehensive understanding of database concurrency control.This paper critiques the ANSI SQL isolation levels, which define isolation based on three phenomena: Dirty Read, Non-Repeatable Read, and Phantom. The authors argue that these definitions are ambiguous and fail to accurately characterize popular isolation levels, including those used in commercial databases. They propose a broader interpretation of these phenomena and introduce new ones to better define isolation types. An important multiversion isolation level, Snapshot Isolation, is defined, which avoids the ANSI SQL phenomena but is not serializable. Snapshot Isolation provides a reduced-isolation level between READ COMMITTED and REPEATABLE READ and is shown to be more powerful than the ANSI SQL isolation levels. The paper also discusses other isolation levels, including Cursor Stability and Snapshot Isolation, and their relationships. It concludes that the ANSI SQL isolation levels are incomplete and that new phenomena are needed to properly define isolation levels. The paper also highlights the importance of considering the broader implications of isolation levels and the need for a more comprehensive understanding of database concurrency control.