This paper presents a handwriting synthesis system developed using hidden semi-Markov model (HSMM) and studies the effect of its parameters namely, number of states and mixture components on the system. A systematic approach is proposed to optimize the number of states in HSMM-based handwriting synthesis system. The method determines shape similarity between generated pattern and average template of the corresponding class. The parameters that provides least difference between the two are chosen as optimized parameters. In addition, a comparison has been made on the selection of optimum number of parameters, required for handwriting recognition system and handwriting synthesis system. All the experiments are carried out on writer dependent Assamese handwritten numeral database and an acceptable result is obtained. © 2016 IEEE.