This paper investigates an unscented attitude estimator based on modified Rodriguez parameter (MRP) and quaternion using a linearized attitude representation transformation. When the singularity transformation conditions do not occur, the developed filter can benefit from the MRP representation in terms of decreased calculation burden and improved filter consistency. When the singularity transformation conditions occur, the developed filter has a comparable performance in terms of accuracy to the well-known unscented quaternion estimator but with better filtering consistency and lower calculation burden. The validation of the developed filter is demonstrated through a high-precision strapdown inertial navigation system/Global Positioning System (SINS/GPS) simulation test and a low-cost SINS/GPS car-mounted experiment.