Abstract: Cropland spatial configuration is an important component of the physical organization of agricultural production, yet the relationship between cropland landscape fragmentation and agricultural total factor productivity (ATFP), as well as the conditions under which this relationship becomes significant, remains insufficiently understood. This study examined that relationship at the county landscape scale using a balanced panel of 749 county-level administrative units in seven typical major grain-producing provinces of China from 2011 to 2023, comprising 9,737 county-year observations. ATFP was estimated with a stochastic frontier analysis based on a translog production function, using primary-industry value added as output and crop sown area, agricultural labor, and total power of agricultural machinery as inputs. Cropland landscape fragmentation was measured from the 30 m China Land Cover Dataset. Eight landscape metrics describing patch size, shape, division, and connectivity were calculated, standardized, and synthesized by principal component analysis to construct a composite cropland fragmentation index. A two-way fixed-effects model with county and year fixed effects was used to estimate the average relationship between landscape fragmentation and ATFP. Robustness was assessed by replacing the fragmentation indicator, replacing the ATFP measure, and using a one-period lag of the fragmentation index. Subgroup regressions were further conducted for northern and southern regions and for areas with low, medium, and high topographic relief. A fixed-effects panel threshold model was then applied to test whether irrigation intensity altered the relationship between fragmentation and ATFP, with the threshold significance evaluated using 300 bootstrap replications. The results showed that cropland landscape fragmentation was significantly and negatively associated with ATFP after county and year fixed effects and socioeconomic controls were included. The baseline coefficient of the fragmentation index was −0.101, and the direction of the relationship remained unchanged in the robustness tests. However, this negative relationship was not spatially uniform. It was statistically significant in southern China but insignificant in northern China. Similarly, the coefficient was significant in areas with high topographic relief, whereas no significant relationship was detected in low- and medium-relief areas. These results indicated that the efficiency implications of landscape fragmentation depended strongly on geographical and production conditions rather than representing a universal effect. The threshold analysis provided further evidence of such conditionality. Irrigation intensity exhibited a significant single threshold at 0.199, with a 95% confidence interval of 0.190–0.210 and a bootstrap p-value of 0.017. When irrigation intensity was at or below the threshold, the coefficient of landscape fragmentation was not statistically significant. When irrigation intensity exceeded 0.199, the coefficient became significantly negative (−0.119), suggesting that fragmentation was more strongly associated with lower ATFP where agricultural production relied more heavily on coordinated irrigation infrastructure and spatially continuous production conditions. Overall, the findings show that, at the county landscape scale, cropland fragmentation is related to agricultural production efficiency in a distinctly context-dependent manner, with the relationship varying across regions, terrain conditions, and irrigation support levels. The results should therefore not be interpreted as direct evidence of farm-level tenure or operational fragmentation mechanisms. Improving agricultural efficiency requires coordinated attention to cropland spatial configuration, farmland infrastructure, and production organization rather than a uniform policy of reducing landscape fragmentation.