The problem with your idea, Jack, is that any decent crypto algorithm will change on average half the output bits when one bit of the key is changed. In other words, by definition, if you can do this, the algorithm isn't as strong as it should be. However, you are correct that failures to meet this standard properly can be found in some popular algorithms. You might want to read Biham and Shamir's book on differential cryptanalysis for details. jack writes:

I got an idea last night, maybe this has already been thought of and tried, but I thought I would give a quick outline of the program I was thinking of:

-Specify a maximum key size (assume 1024bits or something) -Start with an arbitrary key "aaaaaaaaaaaaaa"

Start a loop

-create five mutations of the key -use each key to try and decrypt a few bytes of the message -run a (or some) statistical analysis tests and come up with a value for how 'random' the decrypted bits are -Pick the key that produced the least random ouput

Repeat

Probably this wouldnt work on any very strong algorithm, but it seems it might be effective against some. I am going to write the code and try it out on RC4 and on a weakling little algorithm I wrote a while back. Let me know what yall think.