Online surveys challenge traditional social surveys with several advantages. Due to non-probability sampling and low response rates, however, most online surveys face fundamental flaws. This paper reviews these flaws and limits by exploring key characteristics of online samples. The study employs two sampling designs to collect data. One design follows those who responded to earlier web survey, while the other uses exclusive snowball sampling by personal networks. The resulting “online survey follow-up” sample (n=76) and “personal networks” sample (m=107) contribute to a total of 4492 relationships with alters in the email communication during the past 30 days, and 4674 incoming as well as 3504 outgoing one-on-one personal emails during the past 14 days. These three sets of data are joined and analyzed respectively at the individual, relationship, and contact levels. Major findings suggest that the active respondents in online surveys show distinctive attitudes and values toward the use of online communication. To them the messages exchanged online are important. But the transactions are so transient and fleeting that most messages are not worth keeping. The differences between the two samples are significant and unique in that no other sample characteristics show similar patterns, which further indicates the idiosyncratic tendency among the samples collected by online surveys.