Opinion research is frequently carried out through the Internet and a further increase can be expected. The article focuses on the online access panel, in which respondents are previously recruited through non-probability methods. Despite substantial time- and cost-reduction, online access panel research mainly has to cope with limited Internet coverage and self-selection in the recruitment phase of new panel members. The article investigates whether frequently applied weighting procedures, based on poststratification variables and propensity scores, make online access panel data more representative of the general population. To address this issue, the answers to identical questions are compared between an online self-administered survey of previously recruited online access panel respondents and a face-to-face survey of randomly sampled respondents of the general population. Both respondent groups were surveyed at a similar moment in time (2006-2007) in the same geographical region (Flanders, Belgium). The findings reveal many significant differences, regarding sociodemographic characteristics as well as attitudes towards work, politics and immigrants. The results can be explained by both the specific characteristics of the respondent groups and mode effects. Weighting adjustment had only a minor impact on the results and did not eliminate the differences.