In honeybees, the conditioning of the proboscis extension response (PER) has provided a powerful tool to explore the mechanisms underlying olfactory learning and memory. Unfortunately, PER conditioning does not work well for visual stimuli in intact honeybees, and performance is improved only after antennal amputation, thus limiting the analysis of visual learning and multimodal integration. Here, we study visual learning using the PER protocol in harnessed bumblebees, which exhibit high levels of odor learning under restrained conditions. We trained bumblebees in a differential task in which two colors differed in their rewarding values. We recorded learning performance as well as response latency and accuracy. Bumblebees rapidly learned the task and discriminated the colors within the first two trials. However, performance varied between combinations of colors and was higher when blue or violet was associated with a high reward. Overall, accuracy and speed were negatively associated, but both components increased during acquisition. We conclude that PER conditioning is a good tool to study visual learning, using Bombus impatiens as a model, opening new possibilities to analyze the proximate mechanisms of visual learning and memory, as well as the process of multimodal integration and decision-making.