A manually operated air-driven music box which will generate music at a relatively constant tempo as a user moves the music box back and fourth in opposite directions through the air. The music box has a rotatable drum cylinder to activate tone arms. A first propeller on one end of a drive axle and a second propeller on an opposite end of the drive axle rotate the drive axle as the music box is moved through the air. A gear mechanism for the drive axle rotates the drum cylinder in the same direction as the drive axle is rotated by the propellers. The first propeller is constructed to rotate in a counterclockwise direction and the second propeller is constructed to rotate in a clockwise direction.