Understanding and modeling architectural variability is fundamental in product line engineering. Various extensions have been proposed to architecture description languages (ADLs) to deal with variability. Although these extensions are useful, we argue in this paper that decisions need to be treated as first-class citizens for modeling architectural variability. Decisions that have to be taken by different stakeholders during product derivation are an essential source to understand the variability at different levels (e.g., features, architecture, and implementation). We outline a decision-oriented approach to variability modeling and illustrate it with an example from our ongoing research collaboration with Siemens VAI.