The MLTont ontology is presented in this paper. It was developed to gain a better understanding of testing in machine learning (ML) systems. In light of the complexities and expenses related to software testing, especially in machine learning applications, the ontology aims to clarify the concepts about error, failure, and defect, hence enhancing effective communication among stakeholders. Moreover, MLTont proposes for the adaptation of conventional testing procedures to accommodate the particularities of ML systems, aiming to use current ontologies to enhance interoperability and standard reuse.