Nitrous oxide (N₂O) pulse emissions are detected in soils subjected to freeze–thaw cycles in both laboratory and field experiments. However, the mechanisms underlying this phenomenon are poorly understood. In this study, a laboratory incubation experiment that included freeze–thaw cycles (FTC), freezing (F) and control (CK) treatments was performed on three typical Chinese upland soils, namely, fluvo-aquic soil (FS), black soil (BS) and loess soil (LS). A higher similarity in soil properties and bacterial community structure was discovered between FS and LS than between FS and BS or LS and BS, and the bacterial diversity of FS and LS was higher than that of BS. FTC significantly increased the denitrification potential and the proportion of N₂O in the denitrification gas products in FS and LS but decreased the denitrification potential in BS. Accordingly, with the increasing number of freeze–thaw cycles, the bacterial community composition in the FTC treatments in FS and LS diverged from that in CK but changed little in BS. Taxa that responded to FTC or correlated with denitrification potential were identified. Taken together, our results demonstrated that the effects of FTC on N₂O emissions are soil-type-dependent and that the shift in the microbial community structure may contribute to the elevated N₂O emissions.