This methodological case study illustrates differences in data quality resulting from different online panel recruitment methods. Survey data from two B2B online panels were compared to each other and to data collected from a phone-to-Web sample. One of the online panels was developed via targeted solicitations to B2B Web sites (similar to the non-probability sampling most seen in use today), whereas the other was an ongoing online panel that had been originally recruited by telephone using probabilistic sampling. Many of the problems attributed to online panels (such as "straightlining" or "flatlining") were markedly reduced in the online panel using telephone recruitment with probabilistic sampling, and data from that panel were more robust in statistical analyses. We use this case study to illustrate these differences, and present guidelines for building panels that produce high quality data.