Here, I am done - I think this makes sense.

Read this while reviewing the flow chaeplin posted:
The transactions sent from the mixer (C) to the end address is of a set size (his example was 10). This is D.

He can then search the blockchain for possible candidates (value 10) via a script. Which gets us to block 28531 - this matches the 10.00 XC value. This can be linked at blocked 28533 with the mixer C - we already know C (output 9.99999).
This allows him to trace back to 28531 in the blockchain - to find the values that == 10.00 and match to a specific address - in this case: XQdBjeQtH1JGrkd2MWcXbtsRVeKHWZbnqa which is B.

You can then take all the transactions for this address B - review them and find two matching amounts that == 10 which belong to one single address. This address is A. Since this address B has never been used before this transaction - this is easy to do - and - even if it had multiple transactions - they would not all related back to one single point with one single value (10).