Underwater acoustic channel is the most challenging channel in the world due to its time varying and frequency-selective characters and these characters make channel estimation a compulsory task for coherent orthogonal frequency division multiplexing (OFDM) underwater communication. Channel frequency response can be estimated by pilots that are already known to the receiver. Bit error rate (BER) performance can vary a lot according to different type of pilot pattern. Three kinds of pilot (Block, Comb and Scatter) are analyzed and each of them is proved to work well in some certain underwater environment. Underwater channel is demonstrated to be a comb filter. With channel estimation, stopband and passband of the experiment channel are found and analyzed. Channel estimation error is more likely to occur in stopband than in passband, especially when the channel coherence bandwidth is small. Interchannel interference (ICI) can cause large estimation error by increase its noise level. An improved transform domain filter can reduce the estimation SNR significantly.