The sensitivity of 2-day Northern Hemisphere extratropical forecast errors to changes in initial conditions, computed daily over a 4-year period, is examined. The sensitivity is computed using the (dry) adjoint of the Navy Operational Global Atmospheric Prediction System. This diagnostic enables an assessment of where initial errors have had the largest impact on forecast errors and allows for an examination of interseasonal and interannual variations in predictability. Both the forecast error and sensitivity exhibit a large seasonal cycle, with distinct maxima in winter. The monthly mean sensitivity maxima are correlated with regions of baroclinic instability and occur upstream from the forecast error maxima. Interannual variability is also apparent, and results show that the El Nino winter of 1997/98 was anomalously predictable. In contrast, the recent La Nina winters have been relatively unpredictable, especially during January 2000. These sensitivity calculations highlight the significant impact of middle-lower tropospheric initial condition errors on forecast errors, particularly over the North Pacific.