A lot of research efforts have been devoted to improving energy efficiency for wireless sensor networks by exploring distributed data storage and in-network query processing techniques. In this paper, we present a generic two-tier data storage strategy for answering precision-constrained approximate queries in a sensor network. The basic idea is to keep two versions of data in the network. A high-precision version is kept at the sensor node that captures the data while a low-precision version is maintained at the base station. We develop query processing and node refreshment strategies for various types of approximate queries under the two-tier storage. Our extensive experiments show that the two-tier storage strategy outperforms the basic centralized storage scheme by an order of magnitude in terms of network lifetime under various system configurations.