Background: Many loci associated with obesity have been reported in previous genome-wide association studies (GWASs). However, it remains unclear whether variants at all these loci contributed to onset of obesity or whether one or a few variants cause obesity when obesity is a genetically heterogeneous population. Objective: To investigate the genetic architecture of obesity by clustering a population with obesity into clusters using obesity-related factors. Methods: This study was based on the Tohoku Medical Megabank Project Birth and Three-Generation Cohort Study and the Community-Based Cohort Study. As the Step-1, a GWAS with body mass index (BMI) as an outcome was performed for all 48,365 eligible participants. As the Step-2, we then assigned the 13,067/48,365 participants with obesity (BMI ≥ 25 kg/m2) using the k-prototype to 5 clusters. Obesity-related factors (such as age, nutrient intake, physical activity, sleep duration, difference between weight at age 20 and current weight, smoking, alcohol drinking, psychological distress, and birth weight) were used for clustering. Subsequently, participants in each cluster and those with a BMI < 25 kg/m2 were combined, and GWASs were performed according to the 5 clusters. Additionally, a sub-analysis using data from the UK Biobank was conducted to compare the results. Results: The Step-1 detected 18 genes, most of which were reportedly associated with obesity or obesity-related topics in previous studies. The result of Step-2, of the 18 genes detected in Step-1, LINC01741, CRYZL2P-SEC16B, and SEC16B were significantly related to Cluster 2, FTO, PMAIP1, and MC4R to Cluster 3, and BDNF, BDNF-AS, LINC00678, and KIF18A to Clusters 4 and 5. In the sub-analysis, a similar phenomenon was observed in which separate obesity-related genes were detected for each cluster. Conclusions: Our data support the notion that a decreased sample size with increased homogeneity may reveal insights into the genetic architecture of obesity.