Gamelan is a set of musical instruments from Indonesia and played together as an orchestra. In gamelan playing, one of instruments will act as a conductor so the sound of this instrument must be separated from the others. This paper will discuss the gamelan instrument signal separation using Independent Component Analysis ( ICA ) based on Projection Pursuit ( PP ). The results of previous studies showed that a gamelan instrument signal has nongaussian distribution that is suitable with the requirements of the orthogonal projection search using PP. The experiments of signal separation performed on Demung, Saron, and Peking which play as melodies on gamelan playing. The results show that PP-based ICA (MyICA) of instantaneous mixing are 8% and 11% better than FastICA and ShICA respectively. Meanwhile, MyICA of recorded mixing is 12 % better than FastICA and 13% better than ShICA.