The development of low-cost sensors has generated a corresponding movement to integrate them into many different applications. One such application is determining the rotational attitude of an object. Since many of these low-cost sensors are less accurate than their more expensive counterparts, their noisy measurements must be filtered to obtain optimum results. This work describes the development, testing, and evaluation of four filtering algorithms for the nonlinear sounding rocket attitude determination problem. Sun sensor, magnetometer, and rate sensor measurements are simulated. A quaternion formulation is used to avoid singularity problems associated with Euler angles and other three-parameter approaches. Prior to filtering, Gauss-Newton error minimization is used to reduce the six reference vector components to four quaternion components that minimize a quadratic error function. Two of the algorithms are based on the traditional extended Kalman filter (EKF) and two are based on the recently developed unscented Kalman filter (UKF). One of each incorporates rate measurements, while the others rely on differencing quaternions. All incorporate a simplified process model for state propagation allowing the algorithms to be applied to rockets with different physical characteristics, or even to other platforms. Simulated data are used to develop and test the algorithms, and each successfully estimates the attitude motion of the rocket, to varying degrees of accuracy. The UKF-based filter that incorporates rate sensor measurements demonstrates a clear performance advantage over both EKFs and the UKF without rate measurements. This is due to its superior mean and covariance propagation characteristics and the fact that differencing generates noisier rates than measuring. For one sample case, the "pointing accuracy" of the rocket spin axis is improved by approximately 39 percent over the EKF that uses rate measurements and by 40 percent over the UKF without rates. 7