This study focuses on finding the best viewing angles so that accurate reflectance prediction could be achieved in multi-angle measurement of metallic and pearlescent samples. The root mean square error (RMSE) produced by the reconstruction from the first seven Principle component vectors (PCV), the cumulative contribution being 99.92 percent, were analyzed to find the best viewing angles 10°, 65°, 95°, 120°, 145°, 155° and 170° in clock wise direction from the surface for illumination angle of 45°. The error difference resulted by prediction using cubic interpolation and proposed Principle component analysis (PCA) based method were compared. The prediction by PCA based method outperforms the prediction by cubic interpolation. The method can be employed not only to metallic and pearlescent samples but to all non lambertian surfaces.