Abstract: Wireless sensor network is becoming an important field in wireless network, and data compression is a key technique. Most existing data compression algorithms for wireless sensor network give only emphasis on reducing energy consumption, not considering the limited memory of sensor nodes. In this paper, a problem of memory-efficient data compression for wireless sensor network based on wavelet technique is addressed. A virtual grid-based ring topology and an overlapping clustering topology are firstly designed. Employing those two topologies to perform wavelet transform, border effect can be eliminated. Then, two dimensional and three dimensional data compression transmission algorithms are proposed. In those algorithms, the progressively transmitting data units are specified according to wavelet function and the memory of each cluster head. So, the needed memory of each cluster head doesnt depend on the size of sensory data. The proposed algorithms select sensor nodes to transmit data to cluster head based on spatial correlation among sensory data, and thus high compression efficiency is obtained. From the view points of memory, energy consumption and delay, the performance of those algorithms is analyzed. Theoretically and experimentally it is shown that the proposed algorithm doesnt consume much more energy compared with the existing ones. More importantly, it is memory-efficient.