A statistical framework is proposed to control the laughing body of virtual character agent. The framework takes laughter sound intensity as input. Because of the periodicity of laughter body motion, body motion can be considered as the combination of three different frequency signals. During the training step, the framework captures the relation between intensity signal and output signals such as frequency signals. During the synthesis step, the trained framework takes as input intensity signal and works as generator of frequency signals. Generated body motion is used to animate a laughing virtual character agent.