Face matching tests have traditionally been developed to assess human face perception in the neurotypical range, but methods that underlie their development often make it difficult for these measures to be applied in atypical populations (developmental prosopagnosics, super recognizers) due to unadjusted difficulty. We have recently presented the development of the Oxford Face Matching Test, a measure that bases individual item-difficulty on algorithmically derived similarity of presented stimuli. The measure seems useful as it can be given online or in-laboratory, has good discriminability and high test-retest reliability in the neurotypical groups. In addition, it has good validity in separating atypical groups at either of the spectrum ends. In this talk, I examine the stability of the OFMT and other traditionally used measures in atypical groups. On top of the theoretical significance of determining whether reliability of tests is equivalent in atypical population, this is an important question because of the practical concerns of retesting the same participants across different lab groups. Theoretical and practical implications for further test development and data sharing are discussed.