In this paper, we propose a low complexity linear minimum mean square error (LMMSE) channel estimator for orthogonal frequency division multiplexing (OFDM) system. We partition channel autocorrelation matrix in LMMSE estimator into small sub-matrices. The size of sub-matrix is determined by channel coherence bandwidth. Since the proposed estimator uses small submatrices instead of the whole channel autocorrelation matrix, the complexity is significantly reduced in spite of slight performance degradation. The performance of mean square error (MSE) has been evaluated in exponential decay channel model through computer simulation.