Mobile sensing and proximity-based applications require smart devices to find other nodes in vicinity, though it is challenging for a device to find neighbors in an energy eecient manner while running on low duty cycles. Neighbor discovery schemes allow nodes to follow a schedule to become active and send beacons or listen for other active nodes in order to discover each other with a bounded latency. However, a trade-off exists between the energy consumption and the time a node takes to discover neighbors using a given activity schedule. Moreover, energy consumption is not the only bottleneck, as theoretically perfect schedules can result in discovery failures in a real environment. In this paper, we provide an in-depth study on neighbor discovery, by first deening the relation between energy efficiency, discovery latency and the fraction of discovered neighbors. We evaluate existing mechanisms using extensive simulations for up to 100 nodes and testbed implementations for up to 15 nodes, with no synchronization between nodes and using duty cycles as low as 1% and 5%. Moreover, the literature assumes that multiple nodes active simultaneously always result in neighbor discovery, which is not true in practice as this can lead to collisions between the transmitted messages. Our findings reveal such scalability issues in existing schemes, where discovery fails because of collisions between beacons from multiple nodes active at the same time. Therefore, we show that energy efficient discovery schemes do not necessarily result in successful discovery of all neighbors, even when the activity schedules are computed in a deterministic manner.