For the complex human-machine system such as aircraft cockpit, adaptive function allocation can achieve dynamically changing human-machine collaboration relations, making human and machine work more coordinated and improving the overall effectiveness of the system. A method for adaptive function allocation of aircraft cockpit was proposed, which used Bayesian network to predict performance indexes of every possible level of automation under certain situation, and then, based on multi-attribute risk decision-making, all performance indexes were integrated to obtain the most suitable automation level. The example shows that this method has good operability and flexibility, and can provide technical support for the allocation of aircraft cockpit functions.