Abstract: Exploiting the sparseness of the objective system, a Selective Partial Update algorithm based on Delay Estimate (DE-SPU) is proposed to decrease the length of the adaptive filter. The new algorithm firstly estimates the bulk delay to locate the active coefficients using moving window integration. Then it updates a segment of the non-active coefficients periodically and all the active coefficients at each adaptation. Active coefficients are assigned with large update probability to increase the convergence speed. Non-active coefficients achieve low update probability to ensure the tracking capability. The new algorithm effectively eliminates the information redundancy of the parallel structure by estimating the delay and identifying the active coefficients using only one filter. Finally, the simulation results in the context of echo cancellation indicate the advantages of the proposed algorithm.