Little is known about the effects of grassland degradation on the carbohydrate-active enzyme (CAZYme) genes responsible for C cycling. Here we used a metagenomic approach to reveal variation in abundance and composition of CAZyme genes in grassland experiencing a range of degradation severity (i.e., non-, light, moderately, and severely degraded) in two soil layers (0–10 cm, 10–20 cm) in a steppe grassland in northern China. We observed a higher CAZyme abundance in severely degraded grassland compared with the other three degradation severities. Glycoside hydrolase (GH) and glycosyltransferase (GT) were identified as the most abundant gene families. The Mantel test and variation partitioning suggested an interactive effect of degradation severity and soil depth with respect to CAZyme gene composition. Structural equation modeling indicated that total soil carbon, microbial biomass carbon and organic carbon were the three soil characteristics most important to CAZyme abundance, which suggests an interaction between degradation and soil carbon fractions in determining CAZyme gene composition. Both above- and below-ground factors linked to soil organic matter play a central role in determining the abundance of CAZyme gene families.