This paper introduces the inverse-inverse dynamics method for prediction of human movement and applies it to prediction of cycling motions. Inverse-inverse dynamics optimizes a performance criterion by variation of a parameterized movement. First, a musculoskeletal model of cycling is built in the AnyBody Modeling System (AMS). The movement is then parameterized by means of time functions controlling selected degrees-of-freedom (DOF) of the model. Subsequently, the parameters of these functions are optimized to produce an optimum posture or movement according to a user-defined cost function and constraints. The cost function and the constraints typically express performance, comfort, injury risk, fatigue, muscle load, joint forces and other physiological properties derived from the detailed musculoskeletal analysis. A physiology-based cost function that expresses the integral effort over a cycle to predict the motion pattern and crank torque was used. An experiment was conducted on a group of eight highly trained male cyclists to compare experimental observations to the simulation results. The proposed performance criterion predicts realistic crank torque profiles and ankle movement patterns.