A set of numerical simulations is used to investigate the Pacific Ocean circulation north of 20 ¿S, with emphasis on the Kuroshio/Oyashio current system. The primitive equation models used for these simulations have a free surface and realistic geometry that includes the deep marginal seas such as the Sea of Japan. Most of the simulations have 1/8¿ resolution for each variable but range from 1/2¿, 1.5-layer reduced gravity to 1/16¿, six layer with realistic bottom topography. These are used to investigate the dynamics of the Kuroshio/Oyashio current system and to identify the processes that contribute most to the realism of the simulations. This is done by model-data comparisons, by using the modularity of layered ocean models to include/exclude certain dynamical processes, by varying the model geometry and bottom topography, and by varying model parameters such as horizontal grid resolution, layer structure, and eddy viscosity. In comparison with observational data the simulations show that the barotropic mode, at least one internal mode, nonlinearity, high ''horizontal'' resolution (1/8¿ or finer), the regional bottom topography, and the wind forcing are critical for realistic simulations. The first four are important for baroclinic instability (eddy-mean energetics actually show mixed barotropic-baroclinic instability), the wind curl pattern for the formation and basic placement of the current system, and the bottom topography for the distribution of the instability and for influences on the pathways of the mean flow. Both the Hellerman and Rosenstein (1983) (HR) monthly wind stress climatology and 1000-mbar winds from the European Centre for Medium-Range Weather Forecasts (ECMWF) have been used to drive the model. East of about 150 ¿E, they give a mean latitude for the Kuroshio Extension that differs by about 3¿, approximately 34 ¿N for HR, 37 ¿N for ECMWF, and 35 ¿N observed. The subarctic front is the northern boundary of the subtropical gyre. It is associated with the annual and April--September mean zero wind stress curl lines (which are similar), while the Kuroshio Extension is associated with wintertime zero wind stress curl. This means that part of the flow from the Kuroshio must pass north of the Kuroshio Extension and connect with the Oyashio and subarctic front. Realistic routes for this connection are flow through the Sea of Japan, a nonlinear route separated from the east coast of Japan, and bifurcation of the Kuroshio at the Shatsky Rise. In addition, the six-layer simulations show a 3-Sv meridional overturning cell with southward surface flow and northward return flow centered near 400 m depth. Baroclinic instability plays a critical role in coupling the shallow and abyssal layer circulations and in allowing the bottom topography to strongly influence the shallow circulation. By this means the Izu Ridge and Trench and seamounts upstream and downstream of these have profound influence on (1) the mean path of the Kuroshio and its mean meanders south and east of Japan and (2) on separating the northward flow connecting the Kuroshio and the Oyashio/subarctic front from the east coast of Japan. Without the topographic influence the models show an unrealistic northward current along the east coast of Japan. In essence, the topography regulates the location and strength of the baroclinic instability. The baroclinic instability gives eddy-driven deep mean flows that follow the f/h contours (where f is the Coriolis parameter and h is the depth of the water column) of the bottom topography. These abyssal currents then strongly influence the pathway for subtropical gyre flow north of the Kuroshio Extension and steer the mean meanders in the Kuroshio south and east of Japan. This is corroborated by current meter data from the Kuroshio Extension Regional Experiment (World Ocean Circulation Experiment line PCM 7). ¿ American Geophysical Union 1996