Self-training of Kyudo, i.e. Japanese archery, is difficult for novices but learning from experts is costly. We aim to develop a self-training support system for Kyudo that evaluates shooting forms for trainees. The shooting form of Kyudo, called Shaho-Hassetsu, consists of eight steps. To realize the evaluation function, we plan to build a recognition system on Shaho-Hassetsu that splits the shooting form into the eight steps. To make training data for the Shaho-Hassetsu recognition, we collected shooting form data of Kyudo experts by using the Kinect motion tracker and implemented a tagging support system to identify the eight parts. This paper explains how to collect and process the data for the recognition system.