Measuring and controlling productivity of critical activities will have a significant impact on the cost and time performance of a project. The current practice of acquiring productivity data is heavily dependant on manual collection and data entry; hence it is time consuming, costly and difficult. The objective of the present work is to investigate the automation of the activity sampling method for rapid and real time collection of productivity data. Activity recognition or classification is identified as the major challenge in activity sampling. The proposed method uses accelerometers for activity recognition and a framework was developed for construction activity classification using accelerometers. This paper discusses the preliminary investigation phase in lab environment with a single wired triaxial accelerometer attached to the body of the subject performing typical repetitive movements of a mason. The data patterns were distinct for different movements and hence the features extracted can be used in training for classifying various activity categories. Copyright 2010 ASCE.