We present DIP, a data discovery and dissemination protocol for wireless networks. Prior approaches, such as Trickle or SPIN, have overheads that scale linearly with the number of data items. For T items, DIP can identify new items with O(log(T)) packets while maintaining a O(1) detection latency. To achieve this performance in a wide spectrum of network configurations, DIP uses a hybrid approach of randomized scanning and tree-based directed searches. By dynamically selecting which of the two algorithms to use, DIP outperforms both in terms of transmissions and speed. Simulation and testbed experiments show that DIP sends 20-60% fewer packets than existing protocols and can be 200% faster, while only requiring O(log(log(T))) additional state per data item.