Abstract: In speech processing, robust endpoint detection is one of the most important areas of speech processing. The paper first proposes a feature parameter, called subband power spectral entropy (SPSE). This method is combined with the adaptive band selection (ABS) method proposed by Wu et. al .Finally, a novel robust feature parameter, adaptive subband power spectral entropy (ASPSE), is presented to successfully detect endpoints in different background noises. Experimental results indicate that the ASPSE parameter is very effective under different noise conditions with several SNRs. Furthermore, the proposed algorithm outperforms other algorithms.