Abstract-Long Reed-Solomon (RS) codes are desirable for digital communication and storage systems due to their improved error performance, but the high computational complexity of their decoders is a key obstacle to their adoption in practice. As discrete Fourier transforms (DFTs) can evaluate a polynomial at multiple points, efficient DFT algorithms are promising in reducing the computational complexities of syndrome based decoders for long RS codes. In this paper, we first propose partial composite cyclotomic Fourier transforms (CCFTs) and then devise syndrome based decoders for long RS codes over large finite fields based on partial CCFTs. The new decoders based on partial CCFTs achieve a significant saving of computational complexities for long RS codes. Since partial CCFTs have modular and regular structures, the new decoders are suitable for hardware implementations. To further verify and demonstrate the advantages of partial CCFTs, we implement in hardware the syndrome computation block for a (2720, 2550) shortened RS code over GF(2 12 ). In comparison to previous results based on Horner's rule, our hardware implementation not only has a smaller gate count, but also achieves much higher throughputs. Fast algorithms for discrete Fourier transforms (DFTs) over finite fields are promising techniques to overcome this obstacle. This is because all steps except the key equation solver in syndrome-based hard-decision RS decoders [1] -syndrome computation, Chien search, and error magnitude evaluationare polynomial evaluations. Hence, they can be formulated as DFTs over finite fields.Recently, cyclotomic fast Fourier transforms (CFFTs) over finite fields have been used to reduce the complexities of RS decoders [6], [7]. CFFTs proposed in [6], [8], [9] have low multiplicative complexities, but they have very high additive complexities. By using techniques such as the common subexpression elimination (CSE) algorithm in [10], the additive