Objective: To survey potential predictors of success of mandibular advancement device (MAD) therapy in patients with obstructive sleep apnea (OSA), and in particular, to examine anatomical narrowings and sleep-related collapse levels in the upper airway. Methods: This was a prospective study of 62 OSA patients (median apnea–hypopnea index [AHI] of 34), who were treated with a custom-made, monobloc MAD. The upper airway was examined by inspection, nasopharyngoscopy, overnight acoustic reflectometry recording collapses, and cephalometry of soft tissue dimensions (in addition to skeletal parameters). MAD treatment was controlled by polysomnograpy before and after at least five weeks from the beginning of treatment. Independent predictors of actual reduction in AHI and treatment success (reduction in AHI ≥ 50% with residual AHI < 10) were determined, using multivariable linear and logistic regression. Results: Positional OSA (POSA) and nonsupine AHI (adjusted for upper airway narrowness and collapses, together with gender, age, body mass index, neck circumference, and baseline AHI) were the only independent predictors: POSA indicative for success, and nonsupine AHI inversely related to success. Cephalometry was not predictive. Two predictive models were proposed, one based on POSA having a specificity of 70% and sensitivity of 69%, and the other based on nonsupine AHI, generating a receiver operating characteristic (ROC) curve (area under ROC = 0.78). Using the ROC model, specificity could be increased to 80% without lowering sensitivity. Conclusions: Only variables related to sleep position proved to be independent predictors of success with MAD therapy. The results could be explained by the MAD counteracting the mandible from moving backwards when sleeping supine.