"1. INTRODUCTION. The Automatic Reassessment and Relearning System (ARRS) is an extension of the ""Mastery Learning"" Problem sets in the ASSISTment system. ASSISTments system is a freely available web-based tutoring system for 4th through 10th grade mathematics. Mastery Learning is a strategy that requires students to continually work on a problem set until they have achieved a criterion (typically three consecutive correct answers). The idea of ARRS is if a student masters a problem set, the ARRS system will automatically reassess students a week later, a month later, and then finally two months after that. If students fail the re-assessment, they will be given an opportunity to relearn the topic. We used the data from two ARRS experiment classes ran in September and November of 2010, with in total 136 students and 53449 data instances. To simplify the analysis, we focused on the first reassessment and relearning phase, that means only data from the original assignment and the one week later reassessment and relearning phase is considered. 2. METHODOLOGY AND RESULTS. First we analyzed the performance of the standard Knowledge Tracing model on the Automatic Reassessment and Relearning System data, which focus on enhance student long term learning instead of short term knowledge boost. The result shows that lacking of consideration of forgetting causes a significantly over predicting on students’ first opportunity on a new phase of learning. This proves the necessity of modeling forgetting in student long term learning. We tried simple extensions of the standard Knowledge Tracing trying to model forgetting and relearning on new reassessment and relearning phase. The design of the Knowledge Tracing with new phase forgetting model is shown in Figure 1. Fig. 1. The Knowledge Tracing with new phase forgetting model design. The node N indicates if it is a new reassessment and relearning phase, the node K indicates student current knowledge level, the node P indicates the student performance of current question. The parameter P(L0) is the probability of initial knowledge, P(G) is the probability of guess, P(S) is the probability of slip, P(T) is the probability of learning, and P(F) is the probability of forgetting when facing a new learning phase. We did 10 fold cross validation and used the RMSE (Root Mean Squared Error) as a measurement of the predicting accuracy. The new models gave no better result than the standard Knowledge Tracing. Simulation experiment showed a difficulty of the “forgetting” parameter in these new models to converge into its real value. ACKNOWLEDGEMENTS. This research was made possible by the U.S. Department of Education, Institute of Education Science (IES) grants #R305K03140 and #R305A070440, the Office of Naval Research grant # N00014-03-1-0221, NSF CAREER award to Neil Heffernan, and the Spencer Foundation. All the opinions, findings, and conclusions expressed in this article are those of the authors, and do not reflect the views of any of the funders."