Abstract: Automatic approach for motion synthesis is challenging and interesting in character animations. However, previous works indicate that smoothness of transitions and connectivity of behaviours are difficult to achieve simultaneously. We propose a novel framework as the motion thread network (MTN) to address this challenge. An MTN is constructed by taking as input long motion sequences and creating motion threads between similar frames among these sequences, a traversal of the MTN represents a new motion. Two kinds of MTNs are designed: the frame similarity based MTN and the frame distance based MTN. Moreover, we present an implementation for rhythmic motion synthesis and synchronization t o generate realistic and convincing character dance animations using these two MTNs. At last, we analyze the connectivity and smoothness of MTNs compared to motion graphs (MGs) and move trees (MTs) to prove the efficiency.