An exact maximum likelihood error estimation algorithm based on unbiased converted measurements (UCM-EML) was proposed in order to estimate systematic errors of a MMW radar/IR imaging composite system accurately. An error estimation model was formulated based on measurement noises in polar coordinates, then the criterion function and the corresponding negative log likelihood function were given. The algorithm was implemented using a recursive two-step optimization that involves a modified Gauss-Newton procedure. Simulation results show that the UCM-EML algorithm is better than the exact maximum likelihood (EML) algorithm and the modified exact maximum likelihood (MEML) algorithm on performance and convergence rate.