Single Bayesian network usually has bottlenecks improving of distinguishing rate in educational testing. According to studies to improve the distinguishing rate can reach through fusing multiple Bayesian network,especially based on SVM (Support Vector Machine) fused multiple Bayesian network (Yang Zhi Way, 2006) It is the best result to improve the distinguishing rate at present, thus, this research will utilize the inference tool based on SVM fused multiple Bayesian network, using mathematical calculations fast of the computer, establish a computerized adaptiv learning system. Results: 1. To improve the distinguishing rate can reach through fused multiple Bayesian network especially based on SVM which fused multiple Bayesian network is the best. 2. The province topic ratio reaches above 25% in computerized adaptive diagnostic test. 3. The result obviously elevates after computerized adaptive remedial Instruction activities. An Adaptive Learning System based on SVM fused multiple Bayesian Networks can test and teach students in accordance with their aptitude. Key word: Multiple Bayesian Networks , SVM , Computerized Adaptive Diagnostic Test, Computerized Adaptive Remedial Instruction.