Developing highly qualitative software systems is important but at the same time difficult to be achieved due to a constant increase in size and complexity of the systems. Early detection of software defects becomes a must, various methods being researched. The Software Defects Rules Discovery (SDRD) approach aims to address the generation of rules for software defects. The model is based on code metrics values and uses the ant colony system algorithm to discover the best solution. Performed experiments considered one theoretical example and two open-source projects with various setups, in total 22 different configurations being investigated. The obtained results indicated that from the set of considered metrics the ones that provide the best rules are: CBO (Coupling Between Objects), LOC (Lines Of Code), and NPM (Number of Private Methods).