This paper considers the problem of detecting and estimating AM/FM components in the time-frequency plane. It introduces a new algorithm to estimate the ridges corresponding to the instantaneous frequencies of the components, and to segment the time-frequency plane into different `basins of attraction', each basin corresponding to one mode. The technique is based on the structure of the reassignment vector, which is commonly used for sharpening time-frequency representations. Compared with previous approaches, this new method does not need extra parameters, exhibits less sensitivity to the choice of the window and shows better reconstruction performance. Its effectiveness is demonstrated on simulated and real datasets