An eddy-resolving primitive equation general circulation model is used to estimate water-mass subduction rates in the North Atlantic Ocean subtropical gyre. The diagnostics are based on the instantaneous kinematic approach, which allows the calculation of the annual rate of water-mass subduction at a given density range, following isopycnal outcrop positions over the annual cycle. It is shown that water-mass subduction is effected rapidly (∼1–2 months) as the mixed layer depth decreases in spring, consistent with Stommel's hypothesis, and occurs mostly over the area of deep late-winter mixed layers (≥150 m) across the central North Atlantic in the density range 26 ≤ σ ≤ 27.2. Annual subduction rates O(100–200 m yr–1) are found south and east of the Gulf Stream extension in the density range of subtropical mode waters from roughly 26.2 to 26.6. In the northeastern part of the subtropical gyre, annual subduction rates are somewhat larger, O(250 m yr–1), from a density of about 26.9 east of the North Atlantic Current to 27.4 (upper cutoff in this study). The overall basin-integrated subduction rate for subtropical mode waters (26.2 ≤ σ ≤ 26.6) is about 12.2 Sv (Sv ≡ 106 m3 s−1), comparable to the total formation rate inferred from the surface density forcing applied in the model of roughly 11 Sv in this density range. In contrast, basin-integrated rates for denser central water (26.8 ≤ σ ≤ 27.2) provide a vanishingly small net subduction. In this range, eddy correlations (<30 days) between the surface outcrop area and the local subduction rate counteract the net subduction by the mean flow (deduced from monthly averaged model fields). Comparison with estimates of the annual subduction rate based on the annual mean velocity and late-winter mixed layer properties alone, as is usual in climatological and coarse-resolution model analyses, indicates a mismatch of at least 8 Sv in the density range where the model forms subtropical mode water. This mismatch is primarily due to time-varying mixed layer processes rather than small-scale mixing not resolved explicitly by the model. Our diagnostics based on the instantaneous kinematic approach provide a more complete picture of the water-mass formation process than diagnostics based only on air–sea flux or late-winter mixed layer model data. They reveal the crucial importance of both the seasonal mixed layer cycle and mesoscale eddies to the overall formation rate and provide thus a valuable tool for the analysis of water-mass formation rates in eddy-resolving numerical simulations at basin scale.