Genomic surveillance provides a data source complementary to contact tracing to resolve putative transmission chains. However, the role of within-host diversity in transmission is understudied due to a lack of experimental and clinical datasets that capture within-host diversity in both donors and recipients. Here, we assess the utility of deep-sequenced genomic surveillance within a mouse transmission model where the gastrointestinal pathogen Citrobacter rodentium was controllably spread during co-housing of infected and naïve animals. We observed that within-host variants were maintained over multiple transmission steps until fixation or elimination. We present a model for inferring the likelihood that a given pair of samples are linked by transmission, by comparing the allelic frequency at variant genomic loci. Our data affirm that within-host single nucleotide variants (iSNVs) can repeatedly pass from donor to recipient along the transmission chain, and the mere sharing of iSNVs between different transmission pairs offers limited confidence in identifying a transmission pair. Beyond the presence and absence of within-host variants, we show that differences arising in the relative abundance of iSNVs can infer transmission pairs with high precision. An important component of our approach is that the inference is based solely on sequence data, without incorporating epidemiological or demographic data for context. Our model, which substantially reduces the number of comparisons a contact tracer needs to consider, may enhance the accuracy of contact tracing and other epidemiological processes, including early detection of emerging transmission clusters.