If I understood your program correctly, it is not what I was up to. In fact, my final code was very similar to Xia Keshan's (will set his answer to correct shortly). I hope the following code demonstrates, why I did not choose proc sql but a hash table: Data A;   Input Prod Comp;   Datalines; 1 33 1 44 2 11 2 55 2 66 ; Data B;   Input Comp Subcomp;   Datalines; 33 111 33 222 55 666 55 777 55 888 ; Proc SQL;   Select A.Prod, A.Comp, B.Subcomp From A, B Where A.Comp eq B.Comp; Quit; Data Want (Drop=rc);   Length Comp 8. Subcomp 8.;   If _N_ eq 1 Then Do;     Declare Hash H (Dataset:'B',Multidata:'y'); H.Definekey('Comp'); H.Definedata('Subcomp'); H.Definedone(); Call Missing (Comp, Subcomp);   End;   Do While (not Eof);     Set A End=Eof;     rc=H.Find(); If (rc) Then Do;       Call Missing (Subcomp);    Output; End; Do While (not rc);       Output;    rc=H.Find_next(); End;   End; Run; Proc Print Data=Want; Run; ... View more