Introduction: Many studies have reported the association between air pollution and human health based on regulatory air pollution monitoring data. However, because regulatory monitoring networks were not designed for epidemiological studies, the collected data may not provide sufficient spatial contrasts for assessing such associations. Our goal was to develop a monitoring design supplementary to the regulatory monitoring network in Seoul, Korea. This design focused on the selection of 20 new monitoring sites to represent the variability in PM2.5 across people’s residences for cohort studies. Methods: We obtained hourly measurements of PM2.5 at 37 regulatory monitoring sites in 2010 in Seoul, and computed the annual average at each site. We also computed 313 geographic variables representing various pollution sources at the regulatory monitoring sites, 31,097 children’s homes from the Atopy Free School survey, and 412 community service centers in Seoul. These three types of locations represented current, subject, and candidate locations. Using the regulatory monitoring data, we performed forward variable selection and chose five variables most related to PM2.5. Then, k-means clustering was applied to categorize all locations into several groups representing a diversity in the spatial variability of the five selected variables. Finally, we computed the proportion of current to subject location in each cluster, and randomly selected new monitoring sites from candidate sites in the cluster with the minimum proportion until 20 sites were selected. Results: The five selected geographic variables were related to traffic or urbanicity with a cross-validated R2 value of 0.69. Clustering analysis categorized all locations into nine clusters. Finally, one to eight new monitoring sites were selected from five clusters. Discussion: The proposed monitoring design will help future studies determine the locations of new monitoring sites representing spatial variability across residences for epidemiological analyses.