Stroke survivors might lose their walking and balancing abilities, but many studies pointed out that cycling is an effective means for lower limb rehabilitation. However, during cycle training, the unaffected limb tends to compensate for the affected one, which resulted in suboptimal rehabilitation. To address this issue, we present a Virtual Reality-Cycling Training System (VRCTS), which senses the cycling force and speed in real-time, analyzes the acquired data to produce feedback to patients with a controllable VR car in a VR rehabilitation program, and thus specifically trains the affected side. The aim of the study was to verify the functionality of the VRCTS and to verify the results from the ten stroke patients participants and to compare the result of Asymmetry Ratio Index (ARI) between the experimental group and the control group, after their training, by using the bilateral pedal force and force plate to determine any training effect. The results showed that after the VRCTS training in bilateral pedal force it had improved by 0.22 (p = 0.046) and in force plate the stand balance has also improved by 0.29 (p = 0.031); thus both methods show the significant difference.