Experimental evidence shows that an important reason why people tend to imitate others, to exhibit “herd behavior” is that they assume that the others have information that justifies their actions. The information cascade models of Banerjee [1992] and Bikhchandani et al. [1992] are significant developments in showing some general equilibrium and welfare effects of such rational imitative behavior. But these models as specified may be of limited applicability since they assert that differences across groups in herd behavior can be attributed to the random decisions of first movers. Differences across groups in herd behavior might be explained more often in terms of different modes of interpersonal information transmission. Patterns of human conversation imply great selectivity to the kinds of information transmitted within groups.