Hi again Fabio,

here's a more complete example of my solution:

Set Talk Off
t1=Seconds()
Clear
Close Databases All
Close Tables All

On Error ? Message()

* put NN = 200000 or even 1000000!
#Define NN 200

Create Cursor data1(fr i)

For K=1 To NN
   Insert Into data1 Values (Rand()*10)
Next
Index On fr Tag TAGJOIN1

Create Cursor data2 (fr i)
For K=1 To NN
   Insert Into data2 Values (Rand()*10)
Next
Index On fr Tag TAGJOIN2

Create Cursor data3 (fr i)
For K=1 To NN
   Insert Into data3 Values (Rand()*10)
Next
Index On fr Tag TAGJOIN3

Create Cursor table1 (f1 i,f2 i,f3 i,f4 i)
For K=1 To NN
   Insert Into table1 Values (Rand()*10,Rand()*10,Rand()*10,Rand()*10)
Next
Index On f1 Tag rushmore
Set Order To

? "preparation:",Seconds()-t1
t1=Seconds()

Select table1
Set Relation To f2 Into data1
Set Relation To f3 Into data2 Additive
Set Relation To f4 Into data3 Additive
Delete For f1=2 And !Eof("data1") And !Eof("data2") And !Eof("data3") In "table1"

? "delete:",Seconds()-t1
Set Talk On

* SQL equivalent (not supported):
*
* DELETE FROM table1 where f1=2;
*    and exist(select top 1 fr from data1 A where A.fr=table1.f2 order by fr);
*    and exist(select top 1 fr from data2 B where B.fr=table1.f3 order by fr);
*    and exist(select top 1 fr from data3 C where C.fr=table1.f4 order by fr)

Rushmore will optimize f1=2 and the set relations optimize the "joins" of data1 to data3.

Bye, Olaf.