A non-causal robust learning control method is presented for uncertain systems. According to the optimal index of the nominal system, the non-causal learning part of the method is developed, and the robust part is determined by designing robust weightings. Firstly, robust convergence conditions of learning control laws based on the nominal system are derived when specific forms of the robust part are not considered. Then, robust weightings which are related to uncertainties of the system are designed, and the method for designing robust weightings is drawn according to the robust convergence condition. Finally, simulations are implemented for the validating effectiveness of the method, and the conservatism of robust design of different uncertain systems is analyzed.