For decade interest in the analysis of handwritten music, scores have been growing rapidly. It captured the focus in two types recognition of handwritten music scores and writer identification. Many types of research have proposed different algorithms to improve recognition. Staffline removal cannot be considered as a solved problem that too when dealing with ancient music scripts. However, this work proposes to model the problem as a supervised learning classification task. This work proposes a staffline removal method using CNN to evaluate window size and mask size retaining the symbol information with improved accuracy.