In the process of grinding, chatter is the most important reason for the surface of the roll to produce vibration lines, which seriously affects the surface quality of the workpiece. In order to avoid the adverse effects of chatter, an online chatter monitoring method based on time-varying filtered empirical mode decomposition (TVFEMD) and instantaneous energy ratio (IER) is proposed. The method uses reliable indicators to detect the occurrence of chatter in advance, and solves the problem that the chatter characteristics of roll grinding machine are weak in early stage and difficult to identify quickly under background noise. Firstly, the collected vibration signals are processed in real time by sections. Secondly, the time-varying filtering empirical mode decomposition is carried out for signals within each grinding wheel rotation cycle to improve the signal-to-noise ratio. Then, the chatter sensitive frequency band is selected by using the instantaneous frequency and instantaneous energy ratio, and the instantaneous energy ratio of the chatter sensitive frequency band is taken as the chatter characteristic. Finally, the chatter monitoring threshold is determined based on the instantaneous energy ratio rise, and the current machining state is judged. The results show that the proposed method can detect chatter in the transition stage under different processing conditions of roller grinder, and the early warning of chatter can be achieved faster. Compared with traditional timefrequency analysis methods such as EMD, it has obvious advantages in early chatter monitoring.