In this paper, an algorithm is developed to estimate the frequencies of the sinusoids corrupted by white noise. First, the input data is analyzed by a large order autoregressive (AR) spectral estimator, and then we root the prediction error filter (PEF) to get a set of zeros. Using a zero‐clustering algorithm, we partition the zeros into clusters. From these zero clusters, we construct filters to enhance the sinusoids. Then, using frequency estimators, we estimate the frequencies of the sinusoids from the filtered data. The results presented in this paper demonstrate that the proposed algorithm provides a better resolution ability and estimate accuracy than that of the conventional AR spectral estimator, especially for close sinusoids at lowSNR.