Wireless sensor networks usually have limited energy and transmission capacity, and they can’t match the transmission of a great deal of data. So, it is necessary to approximate or aggregate raw data sampled by sensors in networks. By designing an error tree and solving the regression equations set, this paper proposes a data compression scheme with infinite norm error bound for wireless sensor networks. The algorithms in the scheme can simultaneously explore the temporal and multiple-streams correlations among the sensory data. The temporal correlation in one stream is captured by the 1D Haar wavelet transform. For multivariate monitoring sensor networks, some streams from one sensor are selected as the bases according to the correlation coefficient matrix, and the other streams from the same sensor node can be expressed with one of these bases using linear regression. Theoretically and experimentally, it is concluded that the proposed algorithms can effectively exploit the temporal and multiple-streams correlations on the same sensor node and achieve significant data reduction.