An abundance of empirical evidence indicates that spatial ability is significantly correlated with educational performance in science, technology, engineering, and mathematics (STEM) fields. Previous studies have explored the relationship between spatial ability and science achievement in adults. However, the relationship between spatial ability and science achievement in children at different stages of learning remains unclear. This study examined how the relationship between spatial ability and science achievement differed by grade. A total of 29, 524 students, including 17, 378 fourth graders (boys = 9022, girls = 8365, mean age = 9.78) and 12, 137 eighth graders (boys =6294, girls = 5843, mean age =13.73), were selected from the Qingdao Basic Education Quality Assessment database using cluster random sampling. Participants were recruited to complete two basic cognitive processing tasks (three-dimensional mental rotation and nonverbal matrix reasoning), two self-report scales (multiple intelligence and learning persistence), and a curriculum-based science achievement assessment. The study demonstrated that spatial ability and science achievement are closely related in Chinese children. That this relationship was evident in every instance tested shows a certain stability. The relationship was stronger in eighth-grade students than in fourth-grade students, even after controlling for reasoning ability (measured via non-verbal matrix reasoning) and learning persistence. The relationship between spatial ability and science achievement over time depends on how spatial ability is assessed, either through self-reports or cognitive-based tests. This indicates that in children, as they advance in grade and science courses become more difficult, spatial abilities play an increasingly important role in science learning.