Alzheimer's disease (AD) is one of the fastest growing neurological diseases in the world. We evaluate multivariate multiscale sample entropy (mvMSE) and multivariate multiscale permutation entropy (mvMPE) approaches to distinguish resting-state magnetoencephalogram (MEG) signals of 36 AD patients from those of 26 normal controls. We also discuss about choosing the appropriate embedding dimension value as an effective parameter for mvMPE and MPE for the first time. The results illustrate that both the mvMPE and mvMSE can be useful in the diagnosis of AD, although with different running times and abilities. In addition, our findings show that the MEG complexity analysis performed on deeper time scales by mvMPE and mvMSE may be a useful tool to characterize AD. In most scale factors, the average of the mvMPE and mvMSE values of AD patients are lower than those of controls.