A 1993--1996 hindcast of the North Atlantic circulation is presented and analyzed. Multivariate satellite observations (sea surface height and sea surface temperature) are assimilated into an isopycnic coordinate ocean model (Miami Isopycnic Coordinate Ocean Model ) using a reduced order Kalman filter (the SEEK filter). The objective is to validate the assimilation system in hindcast mode before running a near real-time exercise as part of the European DIADEM project. The analysis scheme is upgraded to make the statistical parameterization as realistic as possible, and this is shown to be particularly important as soon as a multivariate data set is used. Second, the paper describes the modifications to the SEEK filter that are necessary to meet the requirements of a layered model and to benefit from the advantages of the isopycnic coordinate. In particular, the need for an initialization procedure is explained, and this dynamical adjustment is shown to be more natural and robust in the context of the isopycnic coordinate. The hindcast experiment is validated by comparisons with observations, by studying innovation statistics, and by comparing the results with independent observations (expendable bathythermograph profiles).