Information on the Web is growing dramatically. Without a recommended system, the users may spend lots of time on the Web in finding the information they are interested in . With the Web becoming the most popular media for collecting, sharing, and distributing information, it is very common for educational institutions, and organizations to develop Web-Based Training (WBT) systems . Data mining in Web log known as Web-log mining or Web mining has been a hot spot of research work. Many Web mining methods based on association rule  have been proposed. Data on the Web is really unstructured, and implementation of association rule has some limitation. Overcome of these limitation can be done with neuro-fuzzy approach but without optimization. In this paper, we present a novel technique for Web-log mining using ART (Adaptive Resonance Network), and compare it with neuro-fuzzy approach.