The emerging diffusion-based molecular communications is a promising scheme for nano-machine communications. Nevertheless, the Brownian motion model, which describes the behavior of molecules, makes the physical channel different from the channel in conventional wireless communications. In particular, the crossovers in time caused by the late arrivals of the molecules severely affect the communication reliability. In traditional communications, channel coding has long been used to enhance the reliability. Through our explorations, it is shown that the commonly used Hamming distance is no longer a good metric for the channel decoding in diffusion-based molecular communications. The conventional concepts of the channel code design cannot be straightforwardly applied. In this paper, the molecular coding (MoCo) distance function has been proposed, and shown to approach the optimum performance beyond the capability of using Hamming distance. This suggests that new paradigms can be developed upon the MoCo distance.