Decisions made at the early design stage have tremendous impacts on building performance (energy consumption, daylight, life cycle cost, natural ventilation, sunshine hours, etc.). Owing to progress in the design process, the opportunity to improve building performance is constantly reducing, while the cost of optimization is constantly increasing. The literature review shows that the commonly used building performance optimization workflow is divided into two categories: the forward optimization workflow and the inverse optimization workflow. In the forward workflow, designers are allowed to optimize building schemes according to feedback gleaned from the performance metrics; in the inverse workflow, however, designers are allowed to utilize software to search for optimal design solutions. Both workflows have their advantages, and their collective advantages can result in a highly efficient building design; however, in practice, the two processes are often separated. Furthermore, the simulation engines used in these two workflows are simulation software quite widely used. Using these software often requires a large amount of information, which are not suitable for an early design. In this paper, a bidirectional workflow for building performance optimization at the early design stage is proposed. The building energy consumption prediction model is then improved to make the workflow provide real-time performance feedback, and the optimization workflow is realized in SketchUp. This approach can provide quick feedback from building performance metrics, and al ows designers to search for optimal solutions, using a genetic algorithm to support early design decisions. Because of the different structures of the simplified model and the standard model in BESTEST, we chose to use the results of DesignBuilder as the baseline to calibrate the simplified model. The model verification results show that the relative deviation of the total energy consumption of working condition 1 and 2 is between 20％ and 27％due to the relatively large heating deviation in Beijing. The relative deviation of the total energy consumption of other cities is within 10％. In future work, we plan to rebuild the codes of the simplified model, and perform energy calibration under the standard procedure in BESTEST. Finally, the workflow is illustrated through a case study. Compared to previous studies, through the inverse-forward workflow and the simplified energy prediction model, the proposed workflow is demonstrated to better provide fast performance optimization at the early design stage.