Abstract: The problem of estimating unintentional frequency modulation on pulse (UFMOP) which is frequently confronted in electronic reconnaissance is considered. A new instantaneous frequency method based on local polynomial approximation of phase (LPAP) is proposed. This method utilizes a sliding time window and estimates the instantaneous frequency of the central time of the analysis window, through approximating the phases of signal samples within the window locally. The approximation process is realized by using polynomial function. To ensure the efficiency of the proposed method, the selected window length should enable a bias-variance tradeoff as good as possible. In this paper, we compare the classical intersection of confidence intervals (ICI) and the Fan and Gijbels’s bandwidth selection (FGBS) window length selection algorithms. Based on the comparison, we recom-mend adopting the FGBS, which is found to be more suitable for UFMOP estimation, to select the window length adaptively for the LPAP. Simulations and real data analysis show that our method outperforms other commonly used instantaneous frequency estimation methods, and verify the conclusion derived from the comparison of window- length-selection algorithms.