We report sample composition discrepancies related to demographic and personality variables occurring in different stages of development of a probability-based online panel. The first stage—selecting eligible participants—produces differences between Internet users and nonusers in age, education, and gender distribution as well as in the personality traits of openness to experience, conscientiousness, and extraversion. The second and third stages of panel development—asking about willingness to participate in the panel and actual participation in online surveys—result in fewer and smaller discrepancies. The results suggest that among the three potential sources of sample composition bias considered, the largest impact comes from coverage differences with regard to Internet access.