Individual differences in mathematical skills emerge early and are influenced by a range of cognitive and environmental factors. One of these is the Home Mathematics Environment (HME), which includes adult-child mathematics talk. Nevertheless, large variations in methods used to investigate and code adult-child mathematics talk exist. We conducted a pre-registered systematic review (N = 108) of the methods used to code adult-child mathematics talk in children aged 8 years or younger. We identified nine different categories of coding methods. A multi-level meta-analysis with studies additionally reporting children's mathematics skills (291 effect sizes) revealed a small but significant correlation between adult-child mathematics talk and children's mathematics skills (r = 0.072). This was significantly moderated by coding category, implying that some aspects of the interaction are more important than others. This review highlights different methods for coding adult-child mathematical talk, enabling researchers to make informed decisions about how to code interactions.