This paper presents a disturbance-rejection method for a modified repetitive control system with a nonlinearity. Taking advantage of stable inversion, an improved equivalent-input-disturbance (EID) estimator that is more relaxed for system design is developed to estimate and cancel out the influence of the disturbance and nonlinearity in the low-frequency domain. The high-frequency influence is filtered owning to the low-pass nature of the linear part of the closed-loop system. To avoid the restrictive commutative condition and choose a Lyapunov function of a more general form, a new design algorithm, which takes into account the relation between the feedback control gains and the observer and improved EID estimator gains, is developed for the nonlinear system. Furthermore, comparisons with the generalized extended-state observer (GESO) and conventional EID methods are conducted. A clear relation between the developed estimator and the GESO is also clarified. Finally, simulations show the effectiveness and the advantage of the developed method.