In this article we present a new isometric input device for multi-fingered grasping in virtual environments. The device was designed to simultaneously assess forces applied by the thumb, index, and middle finger. A mathematical model of grasping, adopted from the analysis of multi-fingered robot hands, was applied to achieve multi-fingered interaction with virtual objects. We used the concept of visual haptic feedback where the user was presented with visual cues to acquire haptic information from the virtual environment. The virtual object corresponded dynamically to the forces and torques applied by the three fingers. The application of the isometric finger device for multi-fingered interaction is demonstrated in four tasks aimed at the rehabilitation of hand function in stroke patients. The tasks include opening the combination lock on a safe, filling and pouring water from a glass, muscle strength training with an elastic torus, and a force tracking task. The training tasks were designed to train patients' grip force coordination and increase muscle strength through repetitive exercises. The presented virtual reality system was evaluated in a group of healthy subjects and two post-stroke patients (early post-stroke and chronic) to obtain overall performance results. The healthy subjects demonstrated consistent performance with the finger device after the first few trials. The two post-stroke patients completed all four tasks, however, with much lower performance scores as compared to healthy subjects. The results of the preliminary assessment suggest that the patients could further improve their performance through virtual reality training.