Background The STOP-Bang questionnaire is a validated screening tool for the identification of surgical patients with obstructive sleep apnea (OSA). A STOP-Bang score ≥ 3 is highly sensitive but only moderately specific. Apnea/hypopnea during sleep can lead to intermittent hypercapnia and may result in serum bicarbonate (HCO3− ) retention. The addition of serum HCO3− level to the STOP-Bang questionnaire may improve its specificity. Methods Four thousand seventy-seven preoperative patients were approached for consent and screened by the STOP-Bang questionnaire. Polysomnography was performed and preoperative HCO3− level was collected in 384 patients. Study participants were randomly assigned to a derivation or validation cohort. Predictive parameters (sensitivity, specificity, positive and negative predictive values) for STOP-Bang score and serum HCO3− level were calculated. Results In the derivation cohort, with a STOP-Bang score ≥ 3, the specificity for all OSA, moderate/severe OSA, and severe OSA was 37.0%, 30.4%, and 27.7%, respectively. HCO3− level of 28 mmol/L was selected as a cutoff for analysis. With the addition of HCO3− level ≥ 28 mmol/L to the STOP-Bang score ≥ 3, the specificity for all OSA, moderate/severe OSA, and severe OSA improved to 85.2%, 81.7%, and 79.7%, respectively. Similar improvement was observed in the validation cohort. Conclusion Serum HCO3− level increases the specificity of STOP-Bang screening in predicting moderate/severe OSA. We propose a two-step screening process. The first step uses a STOP-Bang score to screen patients, and the second step uses serum HCO3− level in those with a STOP-Bang score ≥ 3 for increased specificity.