The massive consumption of caffeine-containing beverages has prompted many studies involving human participants that have obtained caffeine-based increases in liking for a flavor. However, few studies have succeeded in obtaining caffeine-based flavor preference learning in rats. The main aim of the present study was to examine the conditions under which such learning can be detected. Three experiments differed mainly in terms of the base solution to which caffeine was added. Using a base of maltodextrin and saccharin, Experiment 1 found modest increases in flavor preferences in both food- and fluid-restricted rats. Experiment 2 found a strong caffeine-based flavor preference when water, but not saccharin, was used as the base. Whereas the first two experiments used a within-subject design, in which one flavor was paired with caffeine and a second flavor was not, Experiment 3 used a between-subject design with fluid-restricted rats given almond-flavored water containing caffeine in the Paired condition but not in the Unpaired condition; caffeine-based flavor preference learning was again found. In Experiments 1 and 2 post-conditioning exposure to the flavor alone produced a decrease in preference. In summary, the main achievements of this study were to extend the conditions under which caffeine-based flavor preferences can be detected in rats and demonstrate that such learned preferences are subject to extinction.