In this thesis, an extended Kalman filter formulation for attitude estimation using inertial sensors is developed for implementation in a small satellite simulator. Two different attitude parametrizations are analyzed: quaternions and modified Rodrigues parameters, and for each parametrization, a Kalman filter is designed. This thesis also investigates the deterministic and random error sources of sensors measurements. A calibration procedure is proposed and conducted to eliminate the deterministic error sources. The random error is modeled with the Allan variance formulation and it is integrated in the Kalman filter. Simulations and experiments are conducted to validate the each Kalman filter algorithm. The results for the simulations with both attitude parametrizations are found to yield accurate attitude solutions. The advantage of the quaternion approach is that it does not have singularities, while the advantage of the modified Rodrigues parameters approach is that it is simpler since it has one less state. Experimental tests were conducted on a one-degree of freedom air bearing. The results from experimentation indicate that in reality, the filter results in a less accurate attitude solution than predicted by simulation. Suggestions to improve the filter performance are proposed at the end.