Large variability is a common phenomenon, sometimes a concern, in adapted physical activity research. The cause of large variability is multiple: kind of disability, number of disability, and degree of disability to name a few. Because of large variability both within and across individuals, traditional statistical inference methods usually do not function well and findings, even if the methods can be applied, cannot be generalized to other samples. Unfortunately, many researchers in the field of adapted physical activity research are still focusing on the “mean” or “summary” statistics and ignore the rich information hidden in the variability phenomenon (Bates, 1996; Bouffard, 1997; Rikli, 1997). Furthermore, subjects with large variability were sometimes mistakenly treated as “outliers” and were deleted from the final data analysis. After providing a general review on behavioral variability (e.g., systematic vs. unsystematic variability and inter- vs. intra-subject variability), this presentation will first describe commonly used statistical methods (e.g., discrete and continuous methods) in analyzing variability. How to control unsystematic variability, known also as “noise” or “errors,” in adapted physical activity research (e.g., use a single-case or matching/blocking subject design) will then be described in details, with some practical examples. Finally, the latest progresses in measurement (e.g., individual calibration), statistics (e.g., wavelet transformation) and modeling (e.g., agent-based modeling) in dealing with variability will be introduced.