This work describes a decision theoretic approach for the design of image matching techniques. Theoretically optimal matching algorithms are developed from first principles and analyzed using Monte Carlo simulation. Practical methods are then proposed for reducing computation and improving performance when the image data deviates from the ideal model.