Along with the rapid advancement of technology development led to the amount of information available is also increasingly abundant. The aim of this study was to determine how the implementation of information retrieval system in the classification of the journal by using the cosine similarity and K-Nearest Neighbor (KNN). The data used as many as 160 documents with categories such as Physical Sciences and Engineering, Life Science, Health Science, and Social Sciences and Humanities. Construction stage begins with the use of text mining processing, the weighting of each token by using the term frequency-inverse document frequency (TF-IDF), calculate the degree of similarity of each document by using the cosine similarity and classification using k-Nearest Neighbor.Evaluation is done by using the testing documents as much as 20 documents, with a value of k = {37, 41, 43}. Evaluation system shows the level of success in classifying documents on the value of k = 43 with a value precision of 0501. System test results showed that 20 document testing used can be classified according to the actual category.