Decision fusion rules under fading channel in wireless sensor networks are investigated in this paper. Local decisions made by local sensor nodes may be lost or corrupted while transmitted to the fusion center via a fading channel. A series of fusion rules are proposed under the assumption of Rayleigh channel model. Likelihood ratio rule has been shown optimal through theoretical analysis and simulation. However, it consumes system resource and requires good knowledge of local and channel information, which is not easily available in resource-constrained sensor networks. Three sub-optimal alternatives are proposed, which have less computation and information cost. They perform well in their respective SNR (signal-to-noise ratio) range. Finally, it is found that in resource-constrained wireless sensor networks, a tradeoff should be considered among performance, resource cost and computation complexity while choosing the fusion rules.