A statistical technique that highlights a model's skill in reproducing inter- to multi-decadal (IMD) variability has been successfully used to rank and evaluate models from the Coupled Model Intercomparison Project 3 (CMIP3) dataset. This technique is now applied to the recent CMIP5 dataset and in particular to both the 20th century historical simulations and the 30-year decadal hindcast simulations. Models are then ranked by their ability to simulate observed IMD surface temperature variability during the later part of the 20th century. Rankings of the simulations include uncertainty estimates informed by using multiple ensemble members and unforced simulations. The results suggest that certain models are better able to simulate observed variability and thus may be better candidates for use in the decadal prediction simulations of the 21st century.