The temperature variability over multidecadal and longer timescales (e.g.,the cold epochs in the late 15th,17th,and early 19th centuries) is significant and dominant in the millennium-long,large-scale reconstructions and model simulations;however,their temporal patterns in the reconstructed and simulated temperature series are not well understood and require a detailed assessment and comparison.Here,we compare the reconstructed and simulated temperature series for the Northern Hemisphere (NH) at multidecadal and longer-term timescales (＞30 years) by evaluating their covariance,climate sensitivity and amplitude of temperature changes.We found that covariances between different reconstructions or between reconstructions and simulations are generally high for the whole period of 850-1999 CE,due to their similar long-term temporal patterns.However,covariances between different reconstructions or between reconstructions and simulations steadily decline as time series extends further back in time,becoming particularly small during Medieval times.This is related to the large uncetainties in the reconstructions caused by the decreased number of proxy records and sample duplication during the pre-instrumental periods.Reconstructions based solely on tree-ring data show higher skill than multiproxy reconstructions in capturing the amplitude of volcanic cooling simulated by models.Meanwhile,climate models have a shorter recovery (i.e.,lag) in response to the cooling caused by volcanic eruptions and solar activity minima,implying the lack of some important feedback mechanisms between external forcing and internal climate processes in climate models.Amplitudes of temperature variations in the latest published tree-ring reconstructions are comparable to those of the multiproxy reconstructions.We found that the temperature difference between the Medieval Climate Anomaly (950-1250 CE) and the Little Ice Age (1450-1850 CE) is generally larger in proxy-based reconstructions than in model simulations,but the reason is unclear.