Spectrum prediction, as one enabling technique for dynamic spectrum access (DSA), has recently been received ever-increasing research attention. Most related studies (explicitly or implicitly) assume that spectrum prediction is useful for DSA. However, it does't fundamentally clear when spectrum prediction is really beneficial. To address this issue, this paper firstly provides a high-level review of the existing spectrum sensing (SS) based DSA model and then proposes a sensing-plus-prediction (SPP) based DSA model to reap the benefits of both spectrum sensing and spectrum prediction. Theoretical condition on the usefulness of spectrum prediction for DSA is formally obtained in the form of Theorem. Analytical and simulations results validate that the average throughput of the secondary user based on the proposed SPP DSA model is greater than the SS-based DSA model under the identical target detection probability for the primary user.