In this paper, an orbit determination strategy that uses non-recursive batch filter, and unscented Kalman filter methods is presented and utilized for satellite orbit determination. For the orbit determination system, the range, azimuth, and elevation angles of the satellite measured from ground tracking stations are used for observations. An efficient filter initialization algorithm using the Gibbs method is also proposed to provide an initial state estimate. The non-recursive batch filter is applied to improve initial conditions and to provide initial covariance. Finally, the unscented Kalman filter is used for recursive nonlinear estimation of the states of an Earth-orbiting satellite. The effect of measurement noise and observation duration on orbit determination accuracy is also investigated, and presented.