Large-scale increases in upper ocean temperatures are evident in observational records. Several studies have used well-established detection and attribution (D&A) methods to demonstrate that the observed basin-scale temperature changes are consistent with model responses to anthropogenic forcing, and inconsistent with model-based estimates of natural variability. These studies relied on a single observational dataset, and employed results from only 1 or 2 models. Recent identification of systematic instrumental biases in expendable bathythermograph (XBT) data has however led to improved estimates of ocean temperature variability and trends and raised concerns about earlier results. We examine the causes of ocean warming using these improved observational estimates, together with results from a large multi-model archive of externally forced and unforced simulations. The time evolution of upper-ocean temperature changes is similar in the average of three newer observational temperature-change estimates and in the multi-model average of simulations that include the effects of volcanic eruptions. Our D&A analysis systematically examines the sensitivity of results to a variety of model and data processing choices. In all cases, we obtain a positive identification (at the 1% significance level) of an anthropogenic fingerprint in observed upper-ocean temperatures changes, thereby substantially strengthening existing D&A evidence.