Predictive density for a future observation is derived when the given data comes from circular or spherical distributions. Model robustness of these predictive densities for the circular case is exhibited. Predictive intervals corresponding to the highest density regions are derived and an example is given.