We study the problem of estimating the spacecraft attitude angular velocity using the signal of the spacecraft attitude angular sensors such as the IR horizon sensor, the star sensor, and the full attitude global positioning systems (GPS) sensor for the first time. A signal state model is established. Based on the linear least mean squares (LMS) error criterion, recursive Kalman filtering is developed for the direct estimation of angular velocity. Our algorithm requires relatively few computations and relatively small storage quantities. Finally, the new algorithm is applied to a set of physical simulated IR data, and the results have the advantages of smaller time lag, higher accuracy, and perfect smoothness.