The short-time Fourier transform (STFT) is a typical time-frequency 2-D analysis method for nonstationary signals. Its applications in engineering are limited because of its fixed window, which results in low time-frequency resolution. To surmount this disadvantage, a modified short-time Fourier transform (MSTFT) is proposed. The minimum entropy is adopted as a criterion to determine the width of the window, and variable-width windows to accommodate the signal's characteristics are used instead of constant-width windows. Since it is difficult to find the optimal window width, wavelet packet transformation is utilized to find an approximate solution for it. This method is also applied to multi-target imaging for inverse synthetic aperture radar, and satisfactory results are achieved.