When using old versions, the best solutions may be in old documents.
Your particular question is brilliantly answered in Claudios classic
article: "The mystery of rdb$db_key"
(http://www.cvalde.net/document/mysteriousDbKey.htm). In short, you
should do:

delete from a a1
where a1.rdb$db_key in (
select a2.rdb$db_key
from a a2 join a a3 on a2.code=a3.code
and a2.rdb$db_key>A3.rdb$db_key)

I never thought I would recommend using IN (<subselect>), I always
prefer EXISTS. Though Claudio know what he is talking about and it's a
very long time since I used InterBase 5.6, so you better stick to what
is known to be working.

Set

--- In firebird-support@yahoogroups.com, Johannes Pretorius wrote:
> Good day all
>
> We have a problem with a database of ours. It is in interbase 5.6.
>
> We have a table, for this example called A :
>
>
> CREATE TABLE A (
> CODE VARCHAR(6) NOT NULL
> TITLE T_TITLE,
> INITIALS T_INITIALS,
> SURNAME LASTNAME
> )
>
>
> It has an index on the one column as example
>
> CREATE ASCENDING INDEX RD_CODE ON A (CODE)
>
> Now after a few years we have decided to do what we where suppose to
> do from the beginning and that is to ADD a primary key on the table
> on the CODE column
>
> We do this as follows :
>
> ALTER TABLE A
> ADD CONSTRAINT A_PK1
> PRIMARY KEY (CODE)
>
> We then got the error :
>
> KEY VIOLATION ATTEMPT TO STORE DUPLICATE VALUE (VISIBLE TO ACTIVE
> TRANSACTIONS) IN UNIQUE INDEX "RDB$PRIMARY58"
>
> We then deleted all duplicates as follows
>
> delete from a a1 where (select count(code) from a a2 where a1.code =
> a2.code)>1
>
> We then STILL got the same error. So when then Sweeped the database
> and checked Forced Writes is ON. And dit it again but still the SAME
> error.
>
> We don't really know anymore from here on what to do. If we
> run the following query we get NO records that indicate to be
> duplicated.
>
> select * from a a1 where (select count(code) from a a2 where a1.code
> = a2.code)>1
>
>
> If anybody can give us some ideas it will be appreciated.