We propose an effective compression algorithm for three-dimensional (3-D) mesh sequences, based on principal components analysis (PCA). PCA achieves a coding gain by representing a mesh sequence with a small number of singular vectors. We develop a rate model and a distortion model for the singular vectors and propose an adaptive bit allocation scheme to maximize the coding performance. Specifically, we determine the quantizer step size for each singular vector to minimize the distortion, subject to the constraint on the overall bit rate. Simulation results demonstrate that the proposed algorithm provides much better rate-distortion (RD) performance than the conventional PCA coders.