The temperature variability simulated by climate models is generally consistent with that observed in instrumental records at the scale of global averages, but further insight can also be obtained from regional analysis of the marine temperature record. A protocol is developed for comparing model simulations to observations that account for observational noise and missing data. General consistency between Coupled Model Intercomparison Project Phase 5 model simulations and regional sea surface temperature variability is demonstrated at interannual timescales. At interdecadal timescales, however, the variability diagnosed from observations is significantly greater. Discrepancies are greatest at low latitudes, with none of the 41 models showing equal or greater interdecadal variability. The pattern of suppressed variability at longer timescales and smaller spatial scales appears consistent with models generally being too diffusive. Suppressed variability of low-latitude marine temperatures points to underestimation of intrinsic variability and may help explain why few models reproduce the observed temperature trends during the last 15 years.