The purpose of this study was to verify the effectiveness of modified basketball games based on on-the-ball decision-making and off-the-ball movement. Two types of numerically modified basketball games, “3 on 2” and “3 on 3”, were played by two 6th grade elementary PE classes. Twenty-four students participated in the 3 on 2 games, and 28 students in the 3 on 3 games. All games were videotaped, and the Game Performance Assessment Instrument (GPAI) was used for data analysis. This instrument allows counting of the frequency of on-the-ball decision-making and off-the-ball support movement, and assists in judging the performance of students in situations such as shooting, passing, ball-keeping, and supporting, as to whether or not these are appropriate. The main findings are summarized as follows: 1. The number of students who experienced on-the-ball decision-making and off-the-ball support situations, and the average frequencies of these experiences, were higher in the 3 on 2 games than in the 3 on 3 games. 2. In the 3 on 2 games, the numbers of students who performed appropriate passing, ball-keeping, and support were significantly higher than in the 3 on 3 games (p<001). 3. The rates of appropriate shooting, passing, ball-keeping, and support were significantly higher in the 3 on 2 than in the 3 on 3 games (p<01). The 3 on 2 basketball game was shown to be very effective, allowing upper elementary grade students to learn appropriate on-the-ball decision-making and off-the-ball support movement, in comparison with the 3 on 3 game.