Extensive efforts have been invested on the study design and the statistical methodologies for non-inferiority and superiority trial in therapeutics. Many of their notions are helpful when evaluating non-inferiority and superiority testing in diagnostics. However, the performance of a diagnostic device is described in two dimensions: sensitivity and specificity. The non-inferiority and superiority comparing two devices are assessed by paired data. Appropriate statistical methodologies should be adopted to accommodate the paired data when calculating the confidence interval of performance estimates. This talk will discuss the statistical issues associated with the non-inferiority and superiority testing for diagnostic devices. It will also demonstrate pitfalls with examples and discuss statistical methodologies that may serve best in data analysis.