Abstract： In this paper, a kind of accurate and robust online sequential learning algorithm was proposed for single hidden layer feedforward networks. The algorithm is referred to as online sequential discrete Fourier transformextreme learning machine (OSDFTELM). This approach is able to learn data onebyone or chunkbychunk. During the growth of the data, input weights and output weights are adjusted incrementally. The proposed algorithm has a higher degree of accuracy and robustness compared to the approach referred to as online sequentialextreme learning machine (OSELM). Two simulation examples were presented to show the excellent performance of the proposed approach.