A 4-dimensional variational data assimilation system has been used to better define the mean seasonal state of the North Pacific. The synthesis of available observational records and a sophisticated general circulation model produces a dynamically consistent time-varying dataset which exhibits realistic features of the global ocean circulation and requires no artificial sources or sinks for the temperature and salinity fields. The dataset enables us to clarify the water mass formation and movement processes. A sensitivity experiment using our system reveals that the origin of the North Pacific Intermediate Water can be traced back to the Okhotsk and Bering Seas in the subarctic region and to the subtropical Kuroshio region further south, consistent with recent observational findings. This result illustrates that the ocean state derived from our data assimilation has greater information and forecast potential than that obtained from earlier methods.