In wireless sensor networks, the sensor nodes usually use battery power, one of the main design challenges is to obtain long system lifetime. In a sensor network, the nodes usually sense several kinds of information. These kinds of data have different characteristics. In this paper, the sensed data from the sensor nodes are divided into two types; they are emergent data and usual sensed data. And a data fusion layer is proposed. By using the data reduction algorithm, sensor nodes only transfer the emergent data and the changed information of usual sensed data to sink node. And the sink node recruits the full sensed data. Finally, we implement the data reduction algorithm in TinyOS. Experiment results show that the data reduction algorithm can save energy without losing the sensed information.