Screening compounds for human ether-à-go-go-related gene (hERG) channel inhibition is an important component of early stage drug development and assessment. In this study, we developed a high-confidence (p-value < 0.01) hERG prediction model based on a combined two-dimensional (2D) and three-dimensional (3D) modeling approach. We developed a 3D similarity conformation approach (SCA) based on examining a limited fixed number of pairwise 3D similarity scores between a query molecule and a set of known hERG blockers. By combining 3D SCA with 2D similarity ensemble approach (SEA) methods, we achieved a maximum sensitivity in hERG inhibition prediction with an accuracy not achieved by either method separately. The combined model achieved 69% sensitivity and 95% specificity on an independent external data set. Further validation showed that the model correctly picked up documented hERG inhibition or interactions among the Food and Drug Administration- approved drugs with the highest similarity scores-with 18 of 20 correctly identified. The combination of ascertaining 2D and 3D similarity of compounds allowed us to synergistically use 2D fingerprint matching with 3D shape and chemical complementarity matching.