An Expectation-Maximization approach to sensor calibration is presented and applied to three-axis-magnetometer calibration. This approach is different from the existing attitude-independent approaches mainly in how the attitude parameters in the attitude sensor measurement model are handled. The attitude-independent approaches rely on a conversion of the body and reference representations of the Earth's magnetic field vector into an attitude-independent scalar observation based on scalar checking. The Expectation-Maximization (EM) algorithm essentially maximizes the expectation of the complete likelihood function with respect to the probability distribution of the attitude parameters. Comparisons of the EM algorithm and the attitude-independent approach using simulated magnetometer data are presented.