According to the nonstationarity of rotatary machinary viberating signal and its relationship to rotating speed, a novel elaborate time-frequency(TF) analysis based on variable window Short Time Fourier Transformation（STFT）. Firstly, the rotaing speed curve is optimally segmented according to its slope change, and every segment is fitted using quadratic polynomial. Secondly, based on the fitting parameters the shape and length of moving window are designed respectively. Thirdly, the viberation signal is divided at the segmenting time points obtained above-mentioned, then each signal subsegment is processed using STFT with corresponding window shape and length designed. Finally, the whole TF distribution is obtained by concatenating the subsegment TF spectrum. The simulation and experiment results indicate that the STFT with variable window based on instantenous rotaing speed is very adptable to fine time-frequency analysis for rotary machine viberation signal. It has good research value and application prospects.