Retrospective studies are always observational as the outcome is already available at the time of study. Data can be derived from different sources, including veterinary or herd records, databases containing entries from multiple clinics or herds, registries or databases from national or international organizations. Retrospective studies are appropriate when prospective studies are not feasible, because of time, cost, or ethical constraints. They can provide real-world data and are the basis of most cost or burden of disease studies and have been conducted to answer specific research questions, such as the purchase compliance with long-term medication in companion animals. Retrospective cohort studies and case-control studies are common analyses, which try to find associations between a disease or outcome and exposure. However, retrospective studies are vulnerable to several biases, such as information bias, confounding bias, selection bias, and recall or reporting bias, that must be taken into account when designing or interpreting respective studies. Retrospective research is assumed to become more important, with machine learning techniques allowing the analysis of “Big Data,” which are increasingly collected and exceed the capacity of classical analytic methods.