The documentation for Oracle 9.2 and previous versions contains the following paragraph:

“AFTER row triggers are slightly more efficient than BEFORE row triggers. With BEFORE row triggers, affected data blocks must be read (logical read, not physical read) once for the trigger and then again for the triggering statement.“

For some reason, the documentation in Oracle 10g, and 11g changed as follows:

“BEFORE row triggers are slightly more efficient than AFTER row triggers. With AFTER row triggers, affected data blocks must be read (logical read, not physical read) once for the trigger and then again for the triggering statement. Alternatively, with BEFORE row triggers, the data blocks must be read only once for both the triggering statement and the trigger.“

It turns out that the latest documentation for Oracle 10g and 11g is incorrect and the documentation for Oracle 9i and previous releases is correct: AFTER row triggers are more efficient than BEFORE row triggers. This is true for all versions of Oracle. To verify this, use the following test:

create table tmptrig as select 0 as i from dual;create trigger tmp$before_trigger before update on tmptrigfor each row    begin  if :new.i is null then null; end if; end;/update tmptrig set i = i + 1;set autot trace stat/set autot offdrop trigger tmp$before_trigger;create or replace trigger tmp$after_trigger after update on tmptrigfor each row     begin  if :new.i is null then null; end if; end;/update tmptrig set i = i + 1;set autot trace stat/set autot offdrop trigger tmp$after_trigger;drop table tmptrig;