A robust disturbance reduction scheme for linear small delay systems with disturbances of unknown frequencies is presented in this paper. Unlike other methods, the proposed scheme does not require disturbance frequencies to be known. The linear systems modeled in this study are nominally stable and minimum phase systems with relative degree. The control structure is an integration of Astrom's modified Smith predictor and the proposed scheme. The proposed scheme consists of an input disturbance reduction controller (IDRC) and a residual disturbance reduction controller (RDRC). The IDRC using an artificial neural network (ANN) is proposed to reduce unknown load disturbances and modeling uncertainties in stable systems and unstable systems. The ANN can appropriately approximate the product of an inverse time delay and a nonnegative gain in the IDRC. The residual signals including residual disturbances and residual uncertainties are suppressed by the RDRC based on a disturbance observer. Simulation examples are illustrated to show the effectiveness of the proposed robust disturbance reduction scheme for linear delay uncertain systems with periodic or non-periodic unknown load disturbances.