The invention provides a model-free adaptive control method based on control input saturation. Under the condition of considering control input has position and rate saturation, the invention designs an adaptive neural network constraint controller based on observer technology; and in the design process, a dynamic anti-saturation compensator is provided for adjusting reference preset value in real time to ensure control input does not enter a saturation region. The method is characterized by, to begin with, carrying out model transformation on a common affine nonlinear system through a feedback linearization method; and then, designing a neural network observer and a constraint controller for the transformed system, and providing a dynamic anti-saturation algorithm to adjust a reference set value online to enable the input of the controller to be always within a constraint range.