<p>In drug development, exposure–response models are widely used to inform decisions in dose optimization processes. Type I error (T1) due to misspecified models can lead to critical and costly decisions. Therefore, a new approach called: “Randomized-exposure mixture-model analysis with type 1 error control (REMIX)” to account for model misspecification is proposed and compared against the standard approach (STA). A total of 82 simulation-estimation scenarios for a hypothetical antidiabetic drug were investigated. T1 rate and power was tested in presence or absence of model misspecification. Moreover, predictive performance of both approaches and accuracy of the drug-effect parameter estimates was assessed. Precision and accuracy for the drug-effect parameter were compared to the STA parameter estimates for each structural model with the most patients. REMIX outperformed STA regarding T1 rate inflation (21/82, 44/82 for REMIX and STA, respectively) but led to lower power. In a case study with an example of a clear drug effect and no model misspecification, 27 patients were needed for REMIX compared to 17 for STA to reach 80% power. rRMSE and rBias for full REMIX and STA models were similar in most scenarios if a drug effect was present, but REMIX outperformed STA when no real drug effect was present. Precision and accuracy of parameter estimates were similar for REMIX and STA. Application of REMIX in further studies and comparison to other approaches to control T1 are warranted.</p> Graphical Abstract <p></p>