Incompleteness of datasets is one of the important issues in the area of knowledge discovery in medicine. This study proposes a rough set theory (RST)-based knowledge discovery from coronary artery disease (CAD) datasets when there are only small number of objects and contain missing data (incomplete). At first, RST combined with artificial neural network (ANN) is developed to impute the missing data of the datasets. Then, the knowledge that is discovered from imputed datasets is used to evaluate the quality of the imputation. After that, RST is applied to extract rules from the imputed datasets. This will result in a large number of rules. Rule selection based on the quality of extracted rules is investigated. All the evaluation and selection are based on the complete datasets. Finally, the selected small number of rules is evaluated. The discovered selected rules are used as a classifier on the diagnosis of the presence of CAD to demonstrate their good performance. Copyright © 2011 Inderscience Enterprises Ltd.