Similar to soil moisture, soil temperature has a memory of atmospheric anomalies. However, soil temperature memory over China is still largely unclear, especially in observation. In this study, we investigate the spatiotemporal characteristics of soil temperature memory over China using subsurface (10–80 cm) and deep (160–320 cm) soil temperature data for 626 stations during the period of 1981 to 2005. The red noise method is adopted to estimate soil temperature memory. Results show that the soil temperature memory differs spatially and varies with soil depth and season. Influenced by climate regimes, soil temperature memory at all six layers (with depths of 10, 20, 40, 80, 160, and 320 cm) shows a similar spatial pattern dominated by a northwest to southeast gradient, with relatively high values over arid and semiarid areas of northwestern part of China and relatively low values over humid and semihumid areas of southeastern part of China. During all four seasons, memory lengths increase with soil depth. The average memory of subsurface soil over China can last several months, and for soil at 320 cm, it can be 1 year or more. We also find that seasonal and regional differences of soil temperature memory are stronger in deep soil layers than those in subsurface soil layers. Our findings suggest that soil temperature memory can offer potential for improving seasonal climate prediction over northwestern China. In the meanwhile, the limitations of the methods used in this study should be recognized.