Forward modeling will never be perfect. The shape model will never be able to account for every wrinkle in the MLI coating, and similarly, the surface model as represented by the bidirectional reflectance distribution function (BRDF) will never be identical to that found on actual space objects due to space weathering and inherent differences in the same material from lot to lot. The key to making estimation filters work is thus to continually improve the shape and surface models of the space object, and to intelligently and completely account for all remaining error sources in the models and how these manifest in the observations. This paper uses a multiple hypothesis filtering scheme to determine the attitude of a model geosynchronous satellite with possible surface modeling errors to compare the estimates made when each hypothesis is either propagated by an unscented Kalman filter (UKF) or an unscented Schmidt Kalman filter (USKF). It is shown that the USKF, which incorporates the influence of the known modeling errors in so-called consider terms, provides a better estimate and a more realistic uncertainty on the final attitude estimate.