In a previous work we have proposed ant-miner algorithm based on ant colony optimization (ACO) to classify web users. Ant-miner algorithm capes only nominal/ category attributes, the we can't know real value of the attributes. We must discretise continuous attributes in preprocessing step to support this algorithm running well. In this paper we propose enhancement of ACO algarithm to classify web users for web usage mining: web access log, profile user, and transaction data form e-commerce web. Using these data combination facilitate to classify web users in potential web user class and not potential. In this paper we propose a new heuristic method for web user classification in web usage mining. This heuiristic would he implemented in ACO classification algorithms that copes mix attributes (category and continuous attributes). And we use MDL (Minimum desricption lenght) discretisation to handle continuous attributes. MDL allows a more fexible representation of continuous attributes intervals. In order to implement an improved pheromone updating method. In our e=research, we compared proposed heuristic and information teory heuristic (heuristic that used in origin Ant Miner). The results show that our heuristic method is very compotitive in terms of accuracy and produces signifficantly simpler (smaller) rule sets, a desirable result in data mining)