Laughter is an important social signal in human communication. This paper proposes a statistical framework for generating laughter upper body animations. These animations are driven by two types of input signals, namely the acoustic segmentation of laughter as pseudo-phoneme sequence and acoustic features. During the training step, our statistical framework learns the relationship between the laughter human motion and the input signals. During the synthesis step, our trained framework synthesizes automatically natural head and torso animations from the input signals. Objective and subjective evaluations were conducted to validate this framework. The results show that our proposed framework is capable of generating laughing upper body movements.