Abstract: Document clustering is a technique for classifying documents based on similarity levels of objects within documents. Document clustering is also can be applied to Retrieve Information (IR) based on calculation of term frequency-inverse document frequency and the vector space model-cosine similarity. K-means is one method of grouping One powerful partitioning technique but K-means may be trapped in a local optimum because of centroid random selection. In this study, we build an application document clustering and conduct experiments on the final project document of student in Del Institute of Technology. The experimental results showed that, the K-means clustering which is a partition can be optimized using one of the techniques that single linkage hierarchical clustering based cluster variance (variance within and variance between).