In this article, tri-axial accelerometer devices were used to record swimming motion information at the sacrum. Algorithms have been developed to derive stroke characteristics from the tri-axial accelerometer data. A comparative statistical analysis of manually-collected and accelerometer-derived data for swimming is performed using video data as reference sources. The results show that accelerometer-derived stroke characteristics can be as good as or better than manually-collected data. Accelerometer data processing requires less personnel and technical resources than manual collection or video analysis, enabling greater flexibility of usage and quicker delivery of results. The ability to collect performance measures at every training session could provide a useful tool for assessing training loads, intervention effectiveness, or injury recovery.