And the winner is (roll the drums): David Frankenbach!

Here are the result on a smaller record set (500 records):
- David Frankenbach: 20.750 seconds
- Mark Mccasland: 21.611 seconds
- Josh Weiss: 26.779 seconds
- My original program: 130.317 seconds

Rich Addison and Brad St. Germaine idea don't apply to my problem at hand because I need to check for duplicate in the SDF file itself, not with the master table.

With David solution, I estimate that the total time needed to import 42 485 records will be around 30 minutes instead of the 3 hours with my original idea!

I was surprised by Josh result: I didn't think that it would make that difference between a Rushmorized (!!!) LOCATE and a IndexSeek(). Even the Seek() was faster than the LOCATE.

Anyway, thanks to everyone for their much appreciated ideas.