The diffusion sampling process faces a persistent challenge stemming from its incoherence, attributable to varying noise directions across different timesteps. Our Representative Guidance (RepG) offers a new perspective to handle this issue by reformulating the sampling process with a coherent direction towards a representative target.In this formulation, while the classic classifier guidance improves feature discernment by steering the model away from ambiguous features, it fails to provide a favourable representative target since the class label is overly compact and leads to sacrificed diversity and the adversarial generation problem.In contrast, we leverage self-supervised representations as the coherent target and treat sampling as a downstream task, which refines image details and corrects errors rather than settling for simpler samples.Our representative guidance achieves superior performance and illustrates the potential of pre-trained self-supervised models in image sampling. Our findings demonstrate that RepG not only substantially enhances vanilla diffusion sampling but also surpasses state-of-the-art benchmarks when combined with classifier-free guidance.