In this study we used a biofeedback system to evaluate the joint movements of 3-month-old infants in real-time. The computer was set to discriminate a specific leg position as the motor task infants had to discover to receive the reinforcement from a mobile. Two groups of infants were given two different tasks: (1) to cross 85 knee flexion in the flexion group, and (2) to cross 35 knee extension in extension group. The results from this experiment suggest that infants in both groups learned the task; however, they used different motor solutions. Infants in the flexion group demonstrated two distinctly different motor solutions to make the mobile move. One was movement-based and was characterized by an increase in kicking frequency, while the other was posture-based. In contrast, most infants in the extension group only used the movement-based solution to gain the reinforcements. Controlling the knee at the 35 extension positions seems to be more difficult than at the 85 flexion positions for 3-month-old infants. These results indicate that infants are capable of discovering narrowly defined leg motor solutions and that, depending on their individual characteristics and the task demands, they select different motor solutions.