To make students can learn well, it is necessary to group students according to their abilities. This grouping of students aims to allow students to share knowledge based on the diversity of learning performance of each student. Students' academic performance can be influenced by several factors including intellectual, ability to learn and personality of each student. From these factors obtained academic performance results are grouped into five pieces of learning performance that will later be used as a rule base or rule base in the system. Rule base that is used as a benchmark for academic performance of students is (1) very bad, (2) bad, (3) medium, (4) good, and (5) very good. Grouping students based on predicted learning performance using case based reasoning method with nearest neighbor algorithm. The accuracy of the results of learning performance recommendations using Confusion Matrix is 91.82%