Although several studies have shown that ants can recognize chemical cues from their host plants in ant-plant systems, it is poorly demonstrated in ant gardens (AGs). In this interaction, ant species constantly interact with various epiphyte species. Therefore, it is possible to expect a convergence of chemical signals released by plants that could be acting to ensure that ants are able to recognize and defend epiphyte species frequently associated with AGs. In this study, it was hypothesized that ants recognize and differentiate among chemical stimuli released by AG epiphytes and non-AG epiphytes. We experimentally simulated leaf herbivore damage on three epiphyte species restricted to AGs and a locally abundant understory herb, Piper hispidum, in order to quantify the number of recruited Camponotus femoratus (Fabricius) defenders. When exposed to the AG epiphytes Peperomia macrostachya and Codonanthe uleana leaves, it was observed that the recruitment of C. femoratus workers was, on average, respectively 556% and 246% higher than control. However, the number of ants recruited by the AG epiphyte Markea longiflora or by the non-AG plant did not differ from paper pieces. This indicated that ants could discern between chemicals released by different plants, suggesting that ants can select better plants. These results can be explained by evolutionary process acting on both ants’ capability in discerning plants’ chemical compounds (innate attraction) or by ants’ learning based on the epiphyte frequency in AGs (individual experience). To disentangle an innate behavior, a product of classical coevolutionary process, from an ant’s learned behavior, is a complicated but important subject to understand in the evolution of ant-plant mutualisms.