Online surveys of the general public are currently being conducted with two sorts of samples: (1) probability samples, recruited through RDD phone calls or face-to-face visits to respondents' homes, or (2) groups of people who are not selected from the population of interest by probability sampling and instead volunteer to do occasional on-line surveys for money (in response to online ads or email invitations). To compare the quality of data obtained from these two sorts of samples, we have conducted a series of studies in which the same questionnaire was administered simultaneously by an RDD telephone interviewing organization and/or a probability-sample online survey organization plus one or more online survey groups that employ non-probability samples. We compare the results to assess data accuracy and reach surprising conclusions about the effectiveness of various survey data collection methods.