Active automata learning is a technique of querying black box systems and modelling their behaviour. In this paper, we aim to apply active learning in parts. We formalise the conditions on systems—with decomposable input—that make learning in parts possible. The systems are themselves decomposable through non-intersecting subsets of inputs. Learning these subsystems/components requires less time and resources. We prove that the technique works for both two components as well as an arbitrary number of components. We illustrate the usefulness of this technique through a classical example and through a real example from the industry.