Estimates of temporal trends in oceanic anthropogenic carbon dioxide (CO2) rely on the ability of empirical methods to remove the large natural variability of the ocean carbon system. A coupled carbon-climate model is used to evaluate these empirical methods. Both the ?C* and multiple linear regression (MLR) techniques reproduce the predicted increase in dissolved inorganic carbon for the majority of the ocean and have similar average percent errors for decadal differences (24.1% and 25.5%, respectively). However, this study identifies several regions where these methods may introduce errors. Of particular note are mode and deep water formation regions, where changes in air-sea disequilibrium and structure in the MLR residuals introduce errors. These results have significant implications for decadal repeat hydrography programs, indicating the need for subannual sampling in certain regions of the oceans in order to better constrain the natural variability in the system and to robustly estimate the intrusion of anthropogenic CO2.