In this paper, we study the performance of a training-based least square (LS) and linear minimum mean-square-error (LMMSE) channel estimation for both hop-by-hop and multi-hop direct forwarding wireless sensor networks over frequency-selective fading channels. Specifically, to investigate the properties of the channel estimation, we accomplish a theoretical analysis of MSE in terms of various link parameters. From the performance evaluation, we analytically present the effects of the number of hops on the MSE performance for channel estimations in both multi-hop networks. Interesting observations of MSE behaviors under various conditions are discussed, and the receiver complexity and channel equalization performance are also analyzed. Finally, through the computer simulations, the analytical results and detection performance are demonstrated.