In this paper, we develop a technique in which the flexural rigidity distribution of a composite bending beam can be approximately determined by minimizing the error between the measured and simulated data excited by a point source. A practical example shows the technique to be effective.