This paper proposes a new batch processor based on the Unscented Transformation and demonstrates its application to the attitude determination of spacecraft. The goal of this study is to establish a batch processor without any linearization approximation. The modified Rodrigues parameters are used for attitude representation to maximize the advantages of the Unscented Transformation. The newly developed algorithm can be used to estimate the state of a system simultaneously processing all measurements collected for a fixed time span. Simulation tests are conducted to view the performance of this new approach with a gyro and vector sensors assumed as measurements. The results are compared with those from the Bayesian batch filter as well as those of the unscented and extended Kalman filter. The new batch filter shows a comparable performance to that of the Bayesian batch filter.