This dissertation presents research in the development and use of a multi-modal visual and haptic virtual model in higher education. The model, named Chemical Force Feedback (CFF), represents molecular recognition through the example of protein-ligand docking, and enables students to simultaneously see and feel representations of the protein and ligand molecules and their force interactions. The research efforts have been divided between educational research aspects and development of haptic feedback techniques. The CFF model was evaluated in situ through multiple data-collections in a university course on molecular interactions. To isolate possible influences of haptics on learning, half of the students ran CFF with haptics, and the others used the equipment with force feedback disabled. Pre- and post-tests showed a significant learning gain for all students. A particular influence of haptics was found on students reasoning, discovered through an open-ended written probe where students' responses contained elaborate descriptions of the molecular recognition process. Students' interactions with the system were analyzed using customized information visualization tools. Analysis revealed differences between the groups, for example, in their use of visual representations on offer, and in how they moved the ligand molecule. Differences in representational and interactive behaviours showed relationships with aspects of the learning outcomes. The CFF model was improved in an iterative evaluation and development process. A focus was placed on force model design, where one significant challenge was in conveying information from data with large force differences, ranging from very weak interactions to extreme forces generated when atoms collide. Therefore, a History Dependent Transfer Function (HDTF) was designed which adapts the translation of forces derived from the data to output forces according to the properties of the recently derived forces. Evaluation revealed that the HDTF improves the ability to haptically detect features in volumetric data with large force ranges. To further enable force models with high fidelity, an investigation was conducted to determine the perceptual Just Noticeable Difference (JND) in force for detection of interfaces between features in volumetric data. Results showed that JNDs vary depending on the magnitude of the forces in the volume and depending on where in the workspace the data is presented.