Temporal knowledge graph reasoning has become a focal point of research due to its potential to unlock the full utility of temporal knowledge graphs in a wide range of applications. Unlike static knowledge graphs, temporal knowledge graphs incorporate the dimension of time, resulting in more complex models for reasoning tasks. Current temporal models primarily focus on developing intricate structures to capture the interactions between entities and relations in conjunction with time, yielding some promising results. However, the critical role of negative samples, which has been well established in static models for improving performance, has been largely overlooked. In this paper, we investigate the temporal rules inherently characteristic of temporal knowledge graphs and propose a temporal rule-validated negative sampling (RVNS) method that can effectively enhance temporal model performance. By incorporating RVNS into four representative models and evaluating them on four benchmark datasets, we demonstrate consistent performance improvements across both event-based and time interval-based datasets.