Clustering is an unsupervised method to group multiple objects based on the similarity automatically. The quality of clustering accuracy is determined by the number of similar objects in a correct cluster group. The robust preprocessing process and the choice of cluster algorithm can increase the efficiency of clustering. The objective of this study is to observe the most suitable method to cluster document in Bahasa Indonesia. We performed tests on several cluster algorithms such as K-Means, K-Means++ and Agglomerative with various preprocessing stages and collected the accuracy of each algorithm. Clustering experiments were conducted on a corpus containing 100 documents in Bahasa Indonesia with a commonly used preprocessing scenario. Additionally, we also attach our preprocessing stages such as LSA function, TF-IDF function, and LSA / TF-IDF function. We tested various LSA dimension reductions values from 10% to 90%, and the result shows that the best percentage of reduction rates between 50%-80%. The result also indicates that K-Means++ algorithm produces better purity values than other algorithms.