简而言之，truncate不会记录任何内容(因此更快，但不能撤消)，而delete会记录日志(并且可以是更大事务的一部分，将回滚等)。如果您在开发中有不想在表中使用的数据，通常最好截断，因为您不会冒填满事务日志的风险

如果将TRUNCATE包装在事务中，则可以回滚。

请参阅下面的两个参考资料并自行测试:-

http://blog.sqlauthority.com/2007/12/26/sql-server-truncate-cant-be-rolled-back-using-log-files-after-transaction-session-is-closed/

http://sqlblog.com/blogs/kalen_delaney/archive/2010/10/12/tsql-tuesday-11-rolling-back-truncate-table.aspx

截断与删除是SQL面试中最臭名昭著的问题之一。一定要向面试官解释清楚，否则可能会让你丢掉这份工作。问题是没有多少人意识到这一点，所以如果你告诉他们YES Truncate可以回滚，他们很可能会认为答案是错误的。

TRUNCATE是快的，DELETE是慢的。

尽管TRUNCATE没有问责制。

DROP The DROP command removes a table from the database. All the tables' rows, indexes and privileges will also be removed. No DML triggers will be fired. The operation cannot be rolled back. TRUNCATE TRUNCATE removes all rows from a table. The operation cannot be rolled back and no triggers will be fired. As such, TRUNCATE is faster and doesn't use as much undo space as a DELETE. Table level lock will be added when Truncating. DELETE The DELETE command is used to remove rows from a table. A WHERE clause can be used to only remove some rows. If no WHERE condition is specified, all rows will be removed. After performing a DELETE operation you need to COMMIT or ROLLBACK the transaction to make the change permanent or to undo it. Note that this operation will cause all DELETE triggers on the table to fire. Row level lock will be added when deleting.

来自:http://www.orafaq.com/faq/difference_between_truncate_delete_and_drop_commands

都是很好的答案，我必须补充一句:

由于TRUNCATE TABLE是一个DDL(数据定义语言)命令，而不是DML(数据操作语言)命令，删除触发器不会运行。

对原始答案的一个小修正——删除也会产生大量的重做(因为undo本身是由重做保护的)。这可以从autotrace输出中看到:

SQL> delete from t1;

10918 rows deleted.

Elapsed: 00:00:00.58

Execution Plan
----------------------------------------------------------
   0      DELETE STATEMENT Optimizer=FIRST_ROWS (Cost=43 Card=1)
   1    0   DELETE OF 'T1'
   2    1     TABLE ACCESS (FULL) OF 'T1' (TABLE) (Cost=43 Card=1)




Statistics
----------------------------------------------------------
         30  recursive calls
      12118  db block gets
        213  consistent gets
        142  physical reads
    3975328  redo size
        441  bytes sent via SQL*Net to client
        537  bytes received via SQL*Net from client
          4  SQL*Net roundtrips to/from client
          2  sorts (memory)
          0  sorts (disk)
      10918  rows processed