We consider a reduced-rank square-root unscented Kalman filter based on the Cholesky decomposition of the state-error covariance. The performance of this filter is compared with an analogous filter based on the singular value decomposition. We evaluate the performance of these filters for illustrative linear and non-linear systems.