A large ( approximately 10(23) J) multi-decadal globally averaged warming signal in the upper 300 m of the world's oceans was reported roughly a decade ago and is attributed to warming associated with anthropogenic greenhouse gases. The majority of the Earth's total energy uptake during recent decades has occurred in the upper ocean, but the underlying uncertainties in ocean warming are unclear, limiting our ability to assess closure of sea-level budgets, the global radiation imbalance and climate models. For example, several teams have recently produced different multi-year estimates of the annually averaged global integral of upper-ocean heat content anomalies (hereafter OHCA curves) or, equivalently, the thermosteric sea-level rise. Patterns of interannual variability, in particular, differ among methods. Here we examine several sources of uncertainty that contribute to differences among OHCA curves from 1993 to 2008, focusing on the difficulties of correcting biases in expendable bathythermograph (XBT) data. XBT data constitute the majority of the in situ measurements of upper-ocean heat content from 1967 to 2002, and we find that the uncertainty due to choice of XBT bias correction dominates among-method variability in OHCA curves during our 1993-2008 study period. Accounting for multiple sources of uncertainty, a composite of several OHCA curves using different XBT bias corrections still yields a statistically significant linear warming trend for 1993-2008 of 0.64 W m(-2) (calculated for the Earth's entire surface area), with a 90-per-cent confidence interval of 0.53-0.75 W m(-2).