Scratching is a symptom of many dermatological disorders, especially atopic dermatitis. For the development of anti-itch medications, there is a need for objective measures of scratching. Wrist actigraphy (monitoring wrist and hand movements with micro-accelerometers) is a promising method for assessing scratching; however, currently available technology has a limited capacity to discriminate scratching from other similar movements. In this study, we investigated methods to improve the specificity of actigraphy for scratch detection on movement data collected from subjects using the PAM-RL actigraph. A k-means cluster analysis was used to differentiate scratching from walking and restless sleep, which are potential confounds for nighttime scratching. Features used in the analysis include variance, peak frequency, autocorrelation value at one lag, and number of counts above 0.01 g's. The k-means cluster analysis exhibited a high sensitivity (0.900.10) and specificity for walking (0.980.05) and restless sleep (0.880.06), respectively, demonstrating the separability of these activities. This work indicates that the features described here can be used to develop a classifier that discriminates scratch from other activities. The described method of scratch detection shows promise as an objective method for assessing scratching movements in clinical trials and longitudinal studies of scratch.