A major challenge in component mode synthesis (CMS) remains the absence of a robust mode selection algorithm which can provide a measure of accuracy of the reduced model compared to the full model. In order to address this challenge, two concepts have been proposed in tandem: mode selection and error estimation methods. In this paper, we reassess performance of a recently proposed mode selection method and an a posteriori error estimation method, and propose an efficient iterative algorithm that combines these two methods. The flexibility-based component mode synthesis (F-CMS) is used to demonstrate the performance of the above methods. Numerical examples illustrate the effectiveness of the present algorithm.