In wireless networks, bit error rate (BER) estimation provides a critical foundation for many upper-layer protocols, greatly influencing the performance of data communications. Thus, it has become a significant research topic currently. However, the existing error estimating codes ignore the BER distribution in real networks. Hence, the estimation accuracy of these codes is low. Based on the analysis of BER distribution in practical 802.11 wireless networks, we propose to exploit the principle of differentiation to improve the accuracy of BER estimation. The main idea of our design, called DEE (differentiated error estimation), is to insert into each data packet multiple levels of error estimating codes which have different estimation capability. All these error estimating codes are randomly and evenly distributed in each packet. Then, BER is derived by the theoretical relationship between BER and the parity check failure probability. Furthermore, DEE utilizes the characteristics of uneven distribution of BER to optimize the estimation capability of each level of codes. It improves the estimation accuracy of BER which occurs more frequently, thus reducing the average estimation error. We implement and evaluate DEE over a 7-node testbed. The experimental result shows that DEE reduces the estimation error by about 44% on average compared with the recent scheme EEC (error estimating code). When the estimation redundancy is low, DEE can decrease the estimation error by about 68%. Furthermore, DEE has smaller estimation bias than EEC.