The purpose of this paper is to test whether probability samples and non-probability samples of volunteers yield different survey results. Nine survey companies collected data with an identical questionnaire, one via RDD telephone interviewing, one via the Internet with a sample recruited via RDD telephone interviewing, and seven via the Internet with non-probability samples of volunteers. We first compared the demographic distributions of the nine samples with the Current Population Survey (CPS). Some preliminary results suggest that the distributions of the two probability samples are generally comparable to CPS, while the seven non-probability samples tend to underrepresent minority groups, such as Blacks and Hispanics, lower-educated people, etc. We then compared the two probability samples to assess their accuracy via comparisons with government and industry benchmarks when they are available. Some preliminary finding suggests that the RDD Internet sample produced more accurate results compared to the RDD telephone sample, with regard to only those questions that we have benchmarks to compare to. But for some other questions, the two probability samples generated quite different results. We will further explore whether these differences were due to mode effects or other systematic biases. We also compared the two probability samples to the seven non-probability samples in terms of accuracy. While some of the results were comparable between probability samples and non-probability samples, others were found significantly different. For one non-probability sample, most of the results were not comparable to other companies, which might be attributed to the extremely skewed demographic distribution of the sample. To adjust this, we will examine whether data quality of the non-probability samples would be improved by creating weights to match the survey demographics to the nation. Lastly, we will compare whether experiments embedded in the survey yielded comparable results across the platforms.