This paper introduces and analyzes an event-driven field-estimation (EDFE) protocol for wireless sensor networks. In our field estimation scheme the sensors derive their own notion of expected reading. Readings differing from the expected are considered events of interest and trigger a data transmission to the sink. The proposed protocol allows sensor nodes to only transmit the events of interest, suppressing the transmission of less important data and, as a consequence, saving energy. The EDFE protocol allows the sink to synchronize the received data and correctly reconstruct the sampled process. The proposed scheme is evaluated in a realistic scenario using real-site collected data and representative network conditions. It is shown that the EDFE protocol reduces up to 50% the number of packets sent to the sink while keeping low the average reconstruction error. Moreover, it is demonstrated how the parameters can be tuned to increase the robustness of the protocol to network losses.