Prediction formulae for sleep-disordered breathing can be useful for excluding a diagnosis, establishing an a priori probability of having a positive test, and for prioritizing patient testing. In general, prediction models have high sensitivity but low specificity. In a study analyzing the performance of four previously described prediction models, sensitivities ranged from 76% to 96%, specificities ranged from 13% to 54%, while positive predictive values ranged between 69% and 77%. The models were useful in identifying patients with a respiratory disturbance index of > or = 20 so that these patients could undergo alternative diagnostic testing strategies. The Berlin Questionnaire was tested in primary care settings and was able to identify high-risk patients fairly accurately. A regression neural network performed well with a sensitivity of 99%, a specificity of 80%, a positive predictive value of 88%, and a negative predictive value of 98%. In obese snorers, a regression model utilizing daytime arterial O2 saturation and specific respiratory conductance was effective for excluding obstructive sleep apnea (OSA). In congestive heart failure patients, risk factors for central sleep apnea include male gender, atrial fibrillation, age >60 years, and wake time PaCO2 <38 mm Hg. In children, risk factors for sleep apnea include obesity, African-American race, sinus problems, and persistent wheezing. There are also racial anthropomorphic differences in OSA patients, with whites having a tendency towards brachycephaly facial types (reduced anterior-posterior cranial dimension) and African-Americans having a tendency toward leptoproscopic facial types (longer facial height and decreased facial width). Further refinement of prediction formulae will improve diagnostic accuracy.