Abstract Context Model transformations play a fundamental role in Model-Driven Engineering (MDE) as they are used to manipulate models and to transform them between source and target metamodels. However, model transformation programs lack significant support to maintain good quality which is in contrast to established programming paradigms such as object-oriented programming. In order to improve the quality of model transformations, the majority of existing studies suggest manual support for the developers to execute a number of refactoring types on model transformation programs. Other recent studies aimed to automate the refactoring of model transformation programs, mostly focusing on the ATLAS Transformation Language (ATL), by improving mainly few quality metrics using a number of refactoring types. Objective In this paper, we propose a novel set of quality attributes to evaluate refactored ATL programs based on the hierarchical quality model QMOOD. Method We used the proposed quality attributes to guide the selection of the best refactorings to improve ATL programs using multi-objective search. Results We validate our approach on a comprehensive dataset of model transformations. The statistical analysis of our experiments on 30 runs shows that our automated approach recommended useful refactorings based on a benchmark of ATL transformations and compared to random search, mono-objective search formulation, a previous work based on a different formulation of multi-objective search with few quality metrics, and a semi-automated refactoring approach not based on heuristic search. Conclusion All these existing studies did not use our QMOOD adaptation for ATL which confirms the relevance of our quality attributes to guide the search for good refactoring suggestions.