This paper presents a series of observing system simulation experiments (OSSEs) which are intended as a design study for a proposed array of instrumented moorings in the tropical Atlantic Ocean. Fields of TOPEX/Poseidon sea surface height anomalies are subsampled with the goal being reconstruction of the original fields through the use of reduced-space Kalman filter data assimilation at a restricted number of locations. Our approach differs from typical identical and fraternal twin experiments in that real observed data (i.e., TOPEX/Poseidon data) are subsampled and used in place of synthetic data in all phases of the OSSEs. In this way the question of how closely a particular model-generated data set resembles nature is avoided. Several data assimilation runs are performed in order to optimize the location of a limited number of moorings for the proposed Pilot Research Moored Array in the Tropical Atlantic (PIRATA). Results of experiments in which data are assimilated at 2 ¿N, 2 ¿S and the equator and the longitude is systematically varied by 5¿ show that the greatest impact of the assimilated data occurs when the observations are taken between 15 ¿W and 30 ¿W. Next, a more systematic technique is presented which allows us to determine optimal points in an objective fashion by applying a least squares regression approach to reconstruct the errors on a dense array of points from the data misfits at any three selected points. The forecast error structure from the Kalman filter is used in a novel way to assess the optimality of mooring locations. From a large sample of triads of points, the optimal mooring locations are found to be along the equator at 35 ¿W, 20 ¿W, and 10 ¿W. Additional experiments are performed to demonstrate the efficacy of the initial and final PIRATA configurations and the added value that can be expected from PIRATA observations beyond existing expendable bathythermograph observations. ¿ 1998 American Geophysical Union