Two point clouds of the same scene reconstructed by different techniques may differ in scale. A new scale registration algorithm is proposed in the paper to find the scale difference between two point clouds. First, principle components analysis is applied on descriptors of the point cloud to derive different cumulative contribution rates (CCR) in different neighbor regions. The discrete CCR of each point cloud are then fitted by polynomial scale curves (PSC). The unknown scale is formulated as a solution minimizing the scale difference between two PSCs. Finally, the optimal solution is explicitly formulated as one eigenvalue of a companion matrix. Experimental results prove that the proposed algorithm achieve higher accuracy than existing methods.