There are three ways to estimate the strength of a Chess program:

1. Let it play in human tournaments at standard time control.
   It is by far the most accurate way, but it can be really costly, and some strong players don't want (fear ?) to play against silicon.
   Pharaon has not yet played in a tournament. It has played a speed chess match against a strong chess player and achieved a performance of 2350 elo.
2. Make matches against other programs.
   It is cheaper ;-) (computers do not need incentive), and thanks to Auto232, Winboard/XBoard, ICC,... it can be completely automated.
   The SSDF has played thousand of game between machine.
   The result is not really accurate (due to killer books ( == cheating) and to the fact that some programs (WChess) behave much better against humans then against other chess programs).
   I have played hundred of games between Pharaon and GNU Chess 4.0 / Crafty.
   It seems that Pharaon is about 250 elo points stronger then GNU Chess and near the strength Crafty.
3. Use some Test Sets.
   The Louguet Chess Test (LCT) and the BT2630 are two well known set of test positions to estimate easily and quickly the strength of a program.
   The estimation given by such tests is not so bad ... BUT ... it is *really* easy to tune a program to achieve pretty well in some tests. I have studied extensively the different position of the LCT, and I have tried to improve Pharaon behavior.
   The result was incredible : a 200 elo point improvement on the LCT elo rating !!!
   The conclusion is that such tests are really interesting only for programs that have not used the tests during the development of the program.
   Trying to tune a program for a test set can only weaken the actual strength of a program.