Complex simulations are often explained in terms of their features. However, the disconnect between such high-level analysis features and their ad hoc realization in a simulation program impedes effectively conducting exploratory analysis across the structural and representational space of the problem domain. Motivated by the significant role that exploratory analysis plays in computational discovery across a broad range of domains from scientific problem solving to policy analysis, a feature-driven simulation modeling strategy is introduced. The strategy leverages the concept of activity, agent, and feature algebras, allowing flexible run-time as well as design-time composition of agent-based simulation programs in terms of features.