Abrupt climate changes, especially at millennial timescales, have become a key topic in paleoclimatology because of their link with the tipping point theory and their extensive impacts on future climate scenarios. Although best documented in ice-cores and marine sediments, they are also recorded in different archives on the land, among which Chinese loess and speleothem records are of particular interest. These records effectively document orbital-to-millennial monsoon variability during the Pleistocene, but the dominant periodicity revealed by proxy records from these two types of archives is different. Nevertheless, millennial-scale oscillations recorded in Chinese loess and speleothem records seem highly similar over the last several glacial cycles. Such millennial fluctuations can be of varying amplitudes, and either more or less abrupt depending on the type of record. The observed abruptness usually depends on the sedimentation rates and sampling resolution in the loess and speleothem records. However, it can also be refined by comparisons with multiple physicochemical and isotopic proxies. Here, we apply a robust statistical method to detect abrupt changes in loess grain size and speleothem δ18O records. As a first step, we compared two reference records (i.e. the NGRIP ice-core and the Hulu speleothem δ18O records) of abrupt climate change to verify the robustness of our method in detecting the well-recognized abrupt events during the last glaciation. This result allows us, in a second step, to compare two high-resolution loess grain-size stacks (i.e. the CHILOMOS, and the LGS640 datasets) with the Chinese speleothem composite record built from Hulu and Sanbao records. Although visually observed rapid grain-size variations were previously interpreted as representing millennial-scale variations, our statistical analysis confirms that whether these abrupt climate changes can be robustly identified is highly dependent on the time resolution of the studied records. In a third step, we detected abrupt transitions in the MGSQ grain size stack, covering the last 3.6 Myrs. Our results also show that both winter and summer monsoons co-varied at glacial-interglacial to millennial timescales, reflecting a three-stage evolution of increasing intensity: (1) from 3.6 Ma to 2.6 Ma, (2) from 2.6 Ma to 1.2 Ma, and (3) from 1.2 Ma to present. Our results finally show that the identified abrupt climate events can be employed for synchronizing climate records on a global scale.