A number of recent studies (e.g. Shin and Sardeshmukh, Climate Dynamics 2011) have established that accurate representation of regional climate changes around the globe requires accurate representation of both the magnitude and pattern of tropical sea surface temperature (SST) changes. Discrepancies of 20th century climate model simulations with respect to observations are attributable not only to unforced natural variability, i.e. climate noise, but also significantly to errors in the forced climate change signal in tropical SSTs. The models (both CMIP3 and CMIP5) are not only deficient in capturing the spatial patterns of observed tropical SST trends but also systematically underestimate their magnitude. This systematic model tendency to produce unrealistic spatially bland tropical SST responses to radiative forcing changes has ripple effects around the globe through atmospheric teleconnections to regional and local climate changes. It leads to large errors (sometimes even of sign) in reproducing, for instance, observed long-term trends in the North Atlantic and North Pacific oscillations and changes in regional extreme weather statistics associated with those oscillations. A comprehensive intercomparison study of CMIP5 model simulations of the 1871 to 2010 period with observations, and also with large ensembles of AMIP simulations of the same period with prescribed observed SSTs, lend further support to these conclusions. Improving the representation of tropical SST changes is a good example of a cutting-edge challenge in climate change science that needs to be addressed to generate confidence in 21st century climate change projections around the globe. Some potentially promising avenues to address this challenge will also be discussed in the talk.