Herding is common in consumers in terms of decision making，especially when they are purchasing IT products. However, we are rarely aware of the behavior mechanics underlying herding phenomenon. Software product is a typical kind of IT products. Unfortunately, most users cannot make accurate assessment to the value of software independently due to insufficient knowledge of product information. In such case, herd behavior becomes quite typical in software adoption.In order to further explore the herding puzzle in software adoption, we develop a linear model which is based on Multinomial Logit Marketshare model (MNL), and which could effectively describe the three herding effectsnetwork effect, information cascade effect and wordofmouth effect. We have tested the linear model using the panel data from two different software called Compress/Uncompress Tool and Alarm Clock″ under Sina Software Download channel. Empirical results suggest: (1) Three herding mechanisms exist in software download choicenetwork effect, information cascade effect and wordofmouth effect. ″Recent DownloadsTotal Downloads″ and User Rate″ could be the three variables that represent the cause of the three effects respectively. These three variables have significant positive correlation with the market share of either ″Compress/Uncompress Tool or Alarm Clock,statistically proving the effectiveness of the three mechanisms. Information cascade effect suggests that users tend to choose popular software product to keep up with fashion trends. Network effect suggests that users tend to choose widely used software product for increment of shared value. Wordofmouth effect indicates that users are remarkably influenced by decisions of others to minimize cognitive and emotional conflicts because they believe what most people believe. (2) The rank of the importance of the three mechanisms differs in the two software products adoption. Compared with network effect and information cascade effect, wordofmouth effect plays a more significant role in influencing herd behavior. On the other hand, the influence of network effect and information cascade effect largely depends on the characteristics of software. In terms of ″Compress/Uncompress Tool,″ the network effect is more significant, a phenomenon indicating that users care more about compatibility for sake of conformity to others in a wider range when choosing software involving sharing and swapping. For ″Alarm Clock,″ the information cascade effect is more significant over network effect, a phenomenon indicating that users tend to focus on fashion trends when choosing such standalone software.  This paper extends theoretical herding research by revealing the underlying mechanism in online IT software adoption. The results give implications for designing marketing program and ecommerce website.