Abstract: ESPRIT algorithm is a direct DOA estimation method, of which the performance degrades greatly if the signal-to-noise ratio (SNR) is low and signals correlate with each other. A multi-resolution algorithm for DOA estimation is therefore presented, namely the subband-based ESPRIT algorithm (SB-ESPRIT), which takes advantage of the subband decomposition of fullband signals then estimates the subband DOAs via ESPRIT method. A mapping-back manipulation is used to map the estimated subband spatial frequencies onto the fullband ones. The rotational invariance of the subband signals is also proven. Simulation results showed that SB-ESPRIT will not only improve the resolution of eigenstructure-based method but also decouple the coherent signals if the subbands are properly decomposed. The root mean square error (RMSE) of SB-ESPRIT is also lower than that of ESPRIT.