Patent and proprietary medicine vendors (PPMVs) increase access to antibiotics through non-prescription sales in their drug retail outlets. This fosters irrational antibiotic use among people, thus contributing to the growing burden of resistance. Although training programmes on antibiotic use and resistance exist, they have disproportionately targeted health workers in hospital settings. It’s unclear if there is a relationship between such trainings and non-prescription sales of antibiotics among PPMVs which are more embedded in communities. Therefore, a cross-sectional study was conducted to elicit the determinants of non-prescription antibiotic sales among PPMVs in Kano metropolis, Nigeria. Through brainstorming, causal loop diagrams (CLDs) were used to illustrate the dynamics of factors that are responsible for non-prescription antibiotic sales. Multilevel logistic regression model was used to determine the relationship between training on antibiotic use and resistance and non-prescription antibiotic sales, after controlling for potential confounders. We found that two-third (66.70%) of the PPMVs reported that they have sold non-prescribed antibiotics. A total of three CLDs were constructed to illustrate the complex dynamics of the factors that are related to non-prescription antibiotic sales. After controlling for all factors, PPMVs who reported that they had never received any training on antibiotic use and resistance were twice as more likely to sell antibiotic without prescription compared with those who reported that they have ever received such training (OR = 2.07, 95% CI: 1.27–3.37). This finding suggests that there is an association between training on antibiotic use and resistance and non-prescription sales of antibiotics. However, the complex dynamics of the factors should not be ignored as it can have implications for the development of intervention programmes. Multifaceted and multicomponent intervention packages (incorporating trainings on antibiotic use and resistance) that account for the inherent complexity within the system are likely to be more effective for this setting.