Online social network generates an online mapping of socially connected individuals. These interlinked relationship networks enable new forms of marketing activities that take advantage of the embedded interpersonal influence. However, all online friends are not created equal. Despite possible similarity of positions in the network structure, two different friends of an individual may exert significantly different influence. In the context of an online social network, we examine the influence structure on top of the network structure. A Bayesian model with a reversible jump Markov Chain Monte Carlo procedure is proposed to estimate (1) individual susceptibility to social influence from different relationship categories, (2) dyad-level tie strength within each relationship category, and (3) across-category complexity of social influence among online contacts. Our results suggest that there is significant heterogeneity in individual level social influence structure with respect to category complexity and susceptibility. We identify distinctively different patterns with respect to how activity intensity interacts with individual connectedness to determine influence structure across contact categories. These results offer insights into why seemingly similar viral marketing efforts may have markedly different out-comes. The methodology can be useful for managers to improve the effectiveness of their targeted marketing plans.