In Software Product Lines variability refers to the definition and utilization of differences between several products. Feature Diagrams (FD) are a well-known approach to express variability, and can be used to automate the derivation process. Nevertheless, this may be highly complex due to possible interactions between selected features and the artifacts realizing them. Deriving concrete products typically involves the composition of such inter-dependent software artifacts. This paper presents a feature-based composition approach to automatically derive a product architecture from a given feature configuration. The proposed approach relies on the combination of Model-Driven Engineering (MDE) and Aspect-Oriented Modeling (AOM) techniques. We introduce a metamodel to reify each feature as a high-level aspect model. Product derivation is achieved by weaving the set of aspect models corresponding to a particular feature configuration. The weaving strategy is derived from an in-depth cross-analysis of both the feature interactions and the aspect model dependencies.