This paper deals with the theory of centroid tracking of range-Doppler images created with a heterodyne-detection laser radar. A brief description of the system model is presented and the noise statistics of the image pixels are characterized. A center-of-mass centroid algorithm is used to determine the object's image location within the range-Doppler window. The effects of thresholding and frame averaging upon the performance of the centroid estimator are investigated. An analytical approach for setting the threshold is developed which is based on the criterion of making the estimator unbiased. A constant probability of detection algorithm in conjunction with an acceptable bias condition is described. Performance curves and simulation results are presented which support the theory. Center-of-mass centroid tracking performance, using simulated image data and data collected at the Firepond site in Westford, MA, is shown to be in agreement with the theoretical performance.