In this paper, we are concerned with the problem of adaptive dynamic surface error constrained control for a class of nonlinear multiple-input-multiple-output systems with unknown backlash-like hysteresis nonlinearities. By transforming the tracking errors into new virtual error variables which are incorporated into the proposed prescribed performance control strategy, the prescribed steady-state and transient performance can be ensured. Compared with the existing methods, we introduce the prediction error which is generated between the system state and the serial–parallel estimation model to construct the adaptive laws for neural network weights. The proposed prediction error technique can be used to compensate the tracking error, which implies that a higher accuracy of the identified neural network model is achieved. It is shown that the proposed control approach can guarantee that all the signals of the resulting closed-loop systems are bounded and the output tracks the desired trajectory, while the tracking error are confined all times within the prescribed bounds. Finally, a simulation example is provided to confirm the effectiveness of the proposed approach.