Requirements analysis includes a preliminary acquisition step where a global model for the specification of the system and its environment is elaborated. This model, called requirements model, involves concepts that are currently not supported by existing formal specification languages, such as goals to be achieved, agents to be assigned, alternatives to be negotiated, etc. The paper presents an approach to requirements acquisition which is driven by such higher-level concepts. Requirements models are acquired as instances of a conceptual meta-model. The latter can be represented as a graph where each node captures an abstraction such as, e.g., goal, action, agent, entity, or event, and where the edges capture semantic links between such abstractions. Well-formedness properties on nodes and links constrain their instances—that is, elements of requirements models. Requirements acquisition processes then correspond to particular ways of traversing the meta-model graph to acquire appropriate instances of the various nodes and links according to such constraints. Acquisition processes are governed by strategies telling which way to follow systematically in that graph; at each node specific tactics can be used to acquire the corresponding instances. The paper describes a significant portion of the meta-model related to system goals, and one particular acquisition strategy where the meta-model is traversed backwards from such goals. The meta-model and the strategy are illustrated by excerpts of a university library system.