I too tried the CRC-16 example on our website - but it gave me the wrong answer according to:
http://www.joegeluso.com/software/.../ccitt.htm

I determined this is because the example is not augmented - it only does 8 iterations, rather than the 24 in the longhand example shown in the above reference. If I increase the iterations to 24, I get the right answer for a single 'A', but not for more than 1 character. This is probably because I am not storing the 16 augmented bits from the previous input (it was simply using 0000h again).

I haven't the time to do 24 iterations anyway (I'm using it as a firmware 'flashing' verification, so must loop in much less than 250uS including Winbond writing time - 55uS (38.4kBd download speed). Not tried the table method example as I assume it has the same fault. What about Chi's table example? Anyone proved it?

Dave