Mobile devices are getting more and more popular. With the advanced wireless technology, students can assess Internet with their PDAs. This study aims to develop an adaptive e-learning system on PDA. The idea is to distribute learning materials to the students adaptively. The adaptive engine is based on two levels of Bayesian networks. The first level measures the student performance and the second level determines the redistribution of materials when a student needs retaking. We divide the student performance into the study performance and the quiz performance. The quiz performance is measured by a computerized adaptive testing (CAT) while the study performance relates to the learning behavior of a student. Preliminary experiments demonstrated that student performance has been improved via the adaptive system.