Soil organic carbon (SOC) is important, but research on SOC in Mexican semi-arid grasslands is limited. The objective of this research was to estimate SOC and develop an ecological model to predict soil carbon stocks (SCS) in the semi-arid grasslands of northern Mexico. Sites with different plant cover were selected along the Central Valleys region, and soil samples collected at two microsites (bare soil and beneath grass) at two depths (0–15 cm and 15–30 cm) and analysed for SOC. Plant cover, precipitation, temperature, soil texture and elevation were included as predictor variables to create a SCS prediction model through correlation and regression analyses. SOC varied from 0.465% ± 0.04 (mean ± s.e.m.) in low plant cover–sandy loam soil–low rainfall grasslands to 2.77% ± 0.29 in high plant cover–clay loam soil–high rainfall grasslands. The SOC was higher under grass than bare soil at all sites, while most sites showed similar SOC across soil depth. The prediction model integrated plant cover, mean annual precipitation, elevation and soil sand content (P < 0.0001) as explanatory variables, and reasonably predicted SCS (R2 = 0.798) in semi-arid grasslands of northern Mexico. Our model can be used in grasslands with similar vegetation, climate and soil in northern Mexico, although extrapolation requires caution since further validation at different sites is required.